In recent years, Neural networks are increasingly deployed in various fields to learn complex patterns and make accurate predictions. However, designing an effective neural network model is a challenging task that requires careful consideration of various factors, including architecture, optimization method, and regularization technique. This paper aims to comprehensively overview the state-of-the-art artificial neural network (ANN) generation and highlight key challenges and opportunities in machine learning applications. It provides a critical analysis of current neural network model design methodologies, focusing on the strengths and weaknesses of different approaches. Also, it explores the use of different learning approaches, including convolutional neural networks (CNN), deep neural networks (DNN), and recurrent neural networks (RNN) in image recognition, natural language processing, and time series analysis. Besides, it discusses the benefits of choosing the ideal values for the different components of ANN, such as the number of Input/output layers, hidden layers number, activation function type, epochs number, and model type selection, which help improve the model performance and generalization. Furthermore, it identifies some common pitfalls and limitations of existing design methodologies, such as overfitting, lack of interpretability, and computational complexity. Finally, it proposes some directions for future research, such as developing more efficient and interpretable neural network architectures, improving the scalability of training algorithms, and exploring the potential of new paradigms, such as Spiking Neural Networks, quantum neural networks, and neuromorphic computing.

As AI models become more and more common in business and even in our daily lives, it is important to understand what the carbon impact of these models is. Recent papers have shown that this impact can be quite great, i.e., the training of a single high-end model can result in emissions of more than 500t of CO2eq. In this paper we discuss the factors that influence the carbon footprint of AI models, explore what impact different decisions have, and show how the footprint can be reduced. We also examine the footprint of different models to give a guideline on how urgent action is for different organizations.

Periodic travelling waves are observed in various brain activities including visual, motor, language, sleep, and so on. There are several neural field models describing periodic waves assuming nonlocal interaction and, possibly, inhibition, time delay, or some other properties. In this work we study the influence of asymmetric connectivity functions and of time delay on the emergence of periodic waves and on their properties. Nonlinear wave dynamics is studied, including modulated and aperiodic waves. Multiplicity of waves for the same values of parameters is observed. External stimulation in order to restore wave propagation in a damaged tissue is discussed.

Dynamic neural networks (DNN) are types of artificial neural networks (ANN) that are designed to work with sequential data where context in time is important. In contrast to traditional static neural networks that process data in a fixed order, dynamic neural networks use information about past inputs, which is important if dynamic of a certain process is emphasized. They are widely used in natural language processing, speech recognition and time series prediction. In industrial processes, their use is interesting for the prediction of difficult-to-measure process variables. In an industrial process of isomerization, it is crucial to measure the quality attributes affecting the octane number of gasoline. Process analyzers that are commonly used for this purpose are expensive and subject to failures, therefore, in order to achieve continuous production in case of malfunction, mathematical models for estimating the product quality attributes are imposed as a solution. In this paper, mathematical models were developed using dynamic recurrent neural networks (RNN), i.e., their subtype of a long short-term memory (LSTM) architecture. The results of the developed models were compared with the results of several types of other data-driven models developed for an isomerization process, such as multilayer perceptron (MLP) artificial neural networks, support vector machines (SVM) and dynamic polynomial models. The obtained results are satisfactory, which suggests a good possibility of application.

Deep learning (DL) algorithms have recently emerged from machine learning and soft computing techniques. Since then, several deep learning (DL) algorithms have been recently introduced to scientific communities and are applied in various application domains. Today the usage of DL has become essential due to their intelligence, efficient learning, accuracy and robustness in model building. However, in the scientific literature, a comprehensive list of DL algorithms has not been introduced yet. This paper provides a list of the most popular DL algorithms, along with their applications domains.

The traditional standards employed for pain assessment have many limitations. One such limitation is reliability because of inter-observer variability. Therefore, there have been many approaches to automate the task of pain recognition. Recently, deep-learning methods have appeared to solve many challenges, such as feature selection and cases with a small number of data sets. This study provides a systematic review of pain-recognition systems that are based on deep-learning models for the last two years only. Furthermore, it presents the major deep-learning methods that were used in review papers. Finally, it provides a discussion of the challenges and open issues.

One of the most important sources of energy is the sun. Taiwan is located at north 22-25° latitude. Due to its proximity to the equator, it experiences only a small angle of sunlight incidence. Its unique geographical location which can obtain sustainable and stable solar resources. This study takes research on the forecast of solar radiation to maximize the benefits of solar power generation, and develops methods that can predict the future solar radiation pattern to help reduce the costs of solar power generation. This study builds supervised machine learning models, known as deep neural network (DNN) and long short-term memory neural network (LSTM). The hybrid supervised and unsupervised model, namely cluster-based artificial neural network (k-means clustering and fuzzy C-means clustering-based models), was developed. After establishing these models, the study evaluated their prediction results. For different prediction periods, the study selected the best-performing model based on the results and proposed combining them to establish a real-time updated solar radiation forecast system capable of predicting the next 12 hours. The study area covered Kaohsiung, Hualien, and Penghu in Taiwan. Data from ground stations of the Central Weather Administration, collected between 1993 and 2021, as well as the solar angle parameters of each station, were used as input data for the model. The results of this study show that different models have their advantages and disadvantages in predicting different future times. Therefore, the hybrid prediction system can predict future solar radiation more accurately than a single model.

Drug-target affinity (DTA) prediction is crucial for understanding molecular interactions and aiding drug discovery and development. While various computational methods have been proposed for DTA prediction, their predictive accuracy remains limited, failing to delve into the structural nuances of interactions. With increasingly accurate and accessible structure prediction of targets, we developed a novel deep learning model, named S2DTA, to accurately predict DTA by fusing sequence and structural knowledge of drugs, targets, and pockets using heterogeneous models based on graph and semantic networks. Experimental findings underscored that complex feature representations imparted negligible enhancements to the model’s performance. However, the integration of heterogeneous models demonstrably bolstered predictive accuracy. In comparison to three state-of-the-art methodologies, the supremacy of S2DTA became strikingly apparent. It showcased a noteworthy 25.2% reduction in Mean Absolute Error (MAE) and an impressive 20.1% decrease in Root Mean Square Error (RMSE). Furthermore, S2DTA exhibited substantial advancements in other pivotal metrics, including Pearson Correlation Coefficient (PCC), Spearman, Concordance Index (CI), and R2. These metrics experienced remarkable increments of at least 19.6%, 17.5%, 8.1%, and a remarkable 49.4%, respectively. Finally, we conducted interpretability analysis on the effectiveness of S2DTA by bidirectional self-attention mechanism, fully proving that S2DTA is a valuable and accurate tool for predicting DTA. For further exploration, the source data and code repository can be accessed at https://github.com/dldxzx/S2DTA.

Embedding representation models characterize each word as a vector of numbers with a fixed length. These models have been used in tasks involving text classification, such as recommen- dation and question-answer systems. Semantic relationships are words with a relationship between them providing a complete idea to a text. Therefore, it is hypothesized that an embedding model involving semantic relationships will provide better performance for tasks that use them. This paper presents three embedding models based on semantic relations extracted fromWikipedia to classify texts. The synonym, hyponym, and hyperonym semantic relationships were the ones considered in this work since previous experiments have shown that they are the ones that provide the most semantic knowledge. Lexical-syntactic patterns present in the literature were implemented and subsequently applied to the Wikipedia corpus to obtain the semantic relationships present in it. Several semantic relationships are used in different models: synonymy, hyponym-hyperonym, and a combination of the first two. A convolutional neural network was trained for text classification to evaluate the performance of each model. The results obtained were evaluated with the metrics of precision, accuracy, recall, and F1-measure. The best values obtained with the second model were accuracy of 0.79 for the 20-Newsgroup corpus. F1-measure and recall of 0.87 respectively for the Reuters corpus.

Cancer is a major concern for people, and accurately predicting the probability of cancer incidence and mortality is an important research topic. With the development of big data and artificial intelligence technology, a new machine learning model has emerged. Using 72,591 pieces of data, including age, case count, population size, race, gender, site of onset, and year of discovery, we built a machine learning model. Through calculations, we found that the decision tree, random forest, logistic regression, support vector machine, and neural network achieved testing accuracies of 62.11%, 61.68%, 54.53%, 55.72%, and 63.10%, respectively. With the help of this model, scientists and policymakers can accurately predict future cancer incidence and mortality rates through databases, allowing them to make relevant policies (such as timely allocating doctors and medical resources) to better serve the people.

Buildings have started to play a critical role in the stability and resilience of modern smart grids, leading to a refocusing of large scale energy management strategies from the supply side to the consumer side. When the buildings integrate local renewable energy generation in the form of renewable energy resources they become prosumers and this reflects into additional complexity into the operation of the interconnected complex energy systems. A class of methods of modelling the energy consumption patterns of the building have recently emerged as black-box input-output approaches with the ability to capture underlying consumption trends. These make use and require large quantities of quality data produces by non-deterministic processes underlying the energy consumption. We present an application of a class of neural networks, namely deep learning techniques for time series sequence modelling with the goal of accurate and reliable building energy load forecasting. The Recurrent Neural Network implementation uses Long Short-Term Memory layers in increasing density of nodes to quantify prediction accuracy. The case study is illustrated on four university buildings from temperate climates over one year of operation using a reference benchmarking dataset that allows replicable results. The obtained results are discussed in terms of accuracy metrics and computational and network architecture aspects and are considered suitable for further used in future in situ energy management at the building and neighbourhood levels.

In the context of artificial intelligence's pervasive integration across sectors, the rigorous examination of methodologies used in constructing mathematical models is imperative. This study delves into facial recognition models, particularly in their ability to extract features such as age. We address the prevalent practice of training models with unbalanced datasets and its implications for model performance and bias introduction. Utilizing 27,305 facial images, we investigate the influence of dataset imbalance on age prediction based on ethnicity. Employing preprocessing and oversampling techniques, we equalize sample sizes across ethnicities. A Convolutional Neural Network trained on the balanced dataset serves as our baseline, while three additional models undergo training with a 50% sample size reduction for each group to assess accuracy degradation. Our findings reveal a notable decline in accuracy for ethnic groups with reduced representation. The Asian group experiences a 67.94% accuracy deterioration, followed by a 51.60% drop for the Black group, and a 38.46% decline for the White group. These results underscore the sensitivity of certain ethnic groups to underrepresentation, highlighting the nuanced impact on age prediction accuracy. While a 50% reduction in representation does not uniformly result in a 50% accuracy decline, it unquestionably influences predictive accuracy for affected groups.

In the pursuit of energy-efficient spiking neural network (SNN) hardware, synaptic devices lev-eraging emerging memory technologies hold significant promise. This study investigates the ap-plication of the recently proposed HfO2/SiO2-based interface dipole modulation (IDM) memory for synaptic spike timing-dependent plasticity (STDP) learning. Firstly, through pulse measure-ments of IDM metal-oxide-semiconductor (MOS) capacitors, we demonstrate that IDM exhibits an inherently nonlinear and near-symmetric response. Secondly, we discuss the drain current re-sponse of a field-effect transistor (FET) incorporating a multi-stack IDM structure, revealing its nonlinear and asymmetric pulse response, and suggest that the degree of the asymmetry depends on the modulation current ratio. Thirdly, to emulate synaptic STDP behavior, we implement dou-ble-pulse-controlled drain current modulation of IDMFET using a simple bipolar rectangular pulse. Additionally, we propose a double-pulse-controlled synaptic depression that is valuable for optimizing STDP-based unsupervised learning. Integrating the pulse response characteristics of IDMFETs into a two-layer SNN system for synaptic weight updates, we assess training and classification performance on handwritten digits. Our results demonstrate that ID-MFET-based synaptic devices can achieve classification accuracy comparable to previously re-ported simulation-based results.

In this paper we derive a unified perspective for Optimal Transport (OT) theory and Mean Field Control (MFC) theory to analyse the learning process for Neural Networks algorithms in a high-dimensional framework. We consider Mean Field Neural Networks in the context of MFC theory, specifically the mean field formulation of OT theory that allows the development of highly efficient algorithms while providing a powerful tool in the context of explainable Artificial Intelligence.

The emerging use of visualization techniques in pathology and microbiol- ogy has been accelerated by machine learning (ML) approaches towards image preprocessing, classification, and feature extraction in an increasingly complex series of datasets. Modern Convolutional Neural Network (CNN) architectures have developed into an umbrella of vast image reinforcement and recognition methods, including a combined classification-localization of single/multi-object featured images. As a subtype neural network, CNN cre- ates a rapid order of complexity by initially detecting borderlines, edges, and colours in images for dataset construction, eventually capable in mapping intricate objects and conformities. This paper investigates the disparities between Tensorflow object detection APIs, exclusively, Single Shot Detector (SSD) Mobilenet V1 and the Faster RCNN Inception V2 model, to sample computational drawbacks in accuracy-precision vs. real time visualization capabilities. The situation of rapid ML medical image analysis is theoretically framed in regions with limited access to pathology and disease prevention departments (e.g. 3rd world and impoverished countries). Dark field mi- croscopy datasets of an initial 62 XML-JPG annotated training files were processed under Malaria and Syphilis classes. Model trainings were halted as soon as loss values were regularized and converged.

Identification of plants is a challenging task which aims to identify the family, genus, and species level according to morphological features. Automated deep learning-based computer vision algorithms are widely used for identifying plants and can help users to narrow down the possibilities. However, numerous morphological similarities between and within species make the classification difficult. In this paper, we tested a custom convolution neural network (CNN) and vision transformer (ViT) based models using the PyTorch framework to classify plants. We used a large dataset of 88K and 16K images for classifying plants at genus and species levels respectively. Our results show that for classifying plants at the genus level, ViT models perform better compared to CNN-based models ResNet50 and ResNet-RS-420, and other state-of-the-art CNN-based models suggested in previous studies on a similar dataset. The ViT model achieved top accuracy of 83.3% for classifying plants at the genus level. ViT models also perform better for classifying plants at the species level compared to CNN-based models ResNet50 and ResNet-RS-420, with a top accuracy of 92.5%. We show that the correct set of augmentation techniques plays an important role in classification success.

Stripe rust ​​(caused by Puccinia striiformis f. sp. tritici) is one of the most devastating diseases of wheat and causes large-scale epidemics and severe yield loss. Applying fungicides during early epidemic development is crucial to controlling the disease but is often challenged by resource-limited human visual scouting. Deep learning has the potential to process images and videos captured from affordable devices to empower high-throughput phenotyping for early detection of stripe rust for timely application of fungicides and improve control efficiency. Here, we developed RustNet, a neural network-based image classifier, for efficiently monitoring fields for stripe rust. RustNet was built on a ResNet-18 architecture pre-trained with ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) dataset using transfer learning. RGB images and videos of multiple wheat fields with different wheat types (winter and spring wheat), conditions (irrigated and non-irrigated), and locations were acquired using smartphones or unmanned aerial vehicles near the canopy. A semi-automated image labeling approach was conducted to improve labeling efficiency by combining automated machine labeling and human correction. Cross-validations across multiple categories (sensor platforms, wheat types, and locations) achieved Area Under Curve from 0.72 to 0.87. Independent validation on a published dataset from Germany achieved accuracies ranging from 0.79 to 0.86. The visualization of the last convolutional layer of RustNet demonstrated the identification of pixels with stripe rust. RustNet is freely available at https://zzlab.net/RustNet.

Survival models have become popular for credit risk estimation. Most current credit risk survival models use an underlying linear model. This is beneficial in terms of interpretability but is restrictive for real-life applications since it cannot discover hidden nonlinearities and interactions within the data. This study uses discrete time survival models with embedded neural networks as estimators of time to default. This provides flexibility to express nonlinearities and interactions between variables, and hence allows for models with better overall model fit. Additionally, the neural networks are used to estimate Age-Period-Cohort (APC) models so that default risk can be decomposed into time components for loan age (maturity), origination (vintage) and environment (e.g., economic, operational and social effects). These can be built as general models, or as local APC models for specific customer segments. The local APC models reveal special conditions for different customer groups. The corresponding APC identification problem is solved by a combination of regularization and fitting the decomposed environment time risk component to macroeconomic data, since the environmental risk is expected to have a strong relationship with macroeconomic conditions. Our approach is shown to be effective when tested on a large publicly available US mortgage data set. This novel framework can be adapted by practitioners in the financial industry to improve modelling, estimation and assessment of credit risk.

The purpose of this study is to ​​explore the use of a custom Transformer model in brain-computer interfaces (BCIs) that translate the neural activity present when an individual with limited verbal and fine-motor skills attempts to handwrite. As found in previous studies, Transformers have performed better than recurrent neural networks (RNNs) in translation tasks which are similar to its usage here in decoding neural signals into intended handwritten text. Due to the known benefits of a Transformer, the hypothesis was that the Transformer would show promise in the context of a BCI through the recorded metrics. The neural signals of a tetraplegic individual, when they attempted to handwrite, were provided by existing research. Four trials were conducted using data with or without augmentation which the model used when training to separately minimize training loss and validation loss. When comparing the results of a BCI with the implementation of a Transformer model with the original RNN BCI (the original data source), the Transformer model performed less favorably across all four trials. Although the results do not indicate that the Transformer model currently outperforms an RNN BCI, it is important to note that further testing of the model's capabilities (such as training it with a larger and more preferable dataset and/or for longer, comparing training times between the RNN and Transformer, and/or seeing how the Transformer is improved with an offline autocorrect feature) is necessary before determining whether Transformers can enhance communication in this manner.

Different machine learning models (multiple linear regression, vector support machines, artificial neural networks and random forests) are applied to predict the monthly global irradiation (MGI) from different input variables (latitude, longitude and altitude of meteorological station, month, average temperatures, among others) of different areas of Galicia (Spain). The models were trained, validated and queried using data from three stations, and each best machine model was checked in two independent stations. The results obtained confirmed that the best ML methodology is the ANN model which presents the lowest RMSE value in the validation and querying phases 122.6·10kJ/(m2∙day) and 113.6·10kJ/(m2∙day), respectively, and predict conveniently for independent stations, 201.3·10kJ/(m2∙day) and 209.4·10kJ/(m2∙day), respectively. Given the good results obtained, it is convenient to continue with the design of artificial neural networks applied to the analysis of monthly global irradiation.

The cities of the future are places of sustainable mobility and a continuous reduction in the amount of harmful substances entering the surrounding environment. Current emission models are mainly dedicated to fully combustion vehicles and do not fully reflect the actual emissions of vehicles in the fleet, which is becoming increasingly diverse. The increasing share of hybrid and electric vehicles is generating the need to measure emissions and energy consumption from these vehicles and to develop new emission models that are accurate. This problem mainly concerns low-emission zones in cities, where mobility planning should be based on simulation models that are based on a continuously updated database. This work concerns the proposal of a two-dimensional model of emissions from hybrid vehicles using artificial neural networks for low-emission zones. The results can be used to further develop such emission models.

Change Detection (CD) of aerial images refers to identifying and analyzing changes between two or more aerial images of the same location taken at different times. The CD is a highly challenging task due to the need to distinguish relevant changes, such as urban expansion, deforestation, or post-disaster damage assessment, from irrelevant ones, such as light conditions, shadows, and seasonal variations. Many CD papers have recently been published, where most of the papers that proposed a new model contained a comparison between their proposed and state-of-the-art (SOTA) models. While many recent studies propose new deep learning (DL) models for improving CD performance, their comparative analyses are often restricted, lacking comprehensive insights into the proposed models' real-world generalizability, robustness, and performance trade-offs across diverse change characteristics. This paper presents a novel generic framework to systematically benchmark and assess DL-based CD models through three parallel pipelines: 1) cross-testing models on diverse benchmark datasets to evaluate generalization, 2) robustness analysis against different image corruptions, and 3) multi-faceted contour-level analytics evaluating detection sensitivity to change size/complexity. The framework is applied to comparatively evaluate five state-of-the-art DL-based CD models - Changeformer, BIT, Tiny, SNUNet, and CSA-CDGAN. Extensive experiments unveil each model's strengths, limitations and biases, highlighting their relative proficiencies in generalizing across data distributions, resilience to noise corruption, and discriminative capabilities for changes of varying characteristics. The proposed benchmarking framework demonstrates significant potential for guiding the selection of suitable CD models tailored to specific application requirements by comprehensively evaluating their generalizability, robustness, and detection capabilities across diverse real-world scenarios. This systematic evaluation approach can drive future research into developing more robust and versatile CD solutions aligned with practical needs.

The speed-density or flow-density relationship is a fundamental aspect of traffic flow theory. Single-regime models often face challenges in consistently fitting empirical data under diverse traffic conditions. This paper addresses the inaccuracies of single-regime models, attributing them not only to their functional forms but also to data dispersion and sample selection bias. To mitigate these issues, the data is segmented into day and night periods, leading to the creation of separate models for each. Calibration results, conducted on data from two freeways, Tehran-Karaj and Tehran-Qom, reveal distinct challenges, with the former exhibiting high dispersion and the latter showing low dispersion. Surprisingly, single-regime models demonstrate notable effectiveness when applied to distinct day and night data. Further analysis, categorizing data into day and night, results in reduced errors (20-60% improvement in R-Square), emphasizing the potential for enhanced accuracy by considering distinct regimes. Overall, the study underscores the significance of regime-specific considerations for accurate traffic flow modeling. Upon thorough evaluation, the study highlights the superiority of the neural network Multi-layer Perceptron (MLP) over traditional models. The research discusses the implications of incorporating day and night parameters into the models, emphasizing their potential for accuracy enhancement.

An important indicator of human-related pollution in watersheds is dissolved oxygen (DO). The DO is highly dependent on both space and time characteristics of the watershed and is directly linked to eutrophication, which impairs the development of both the aquatic fauna and flora, also negatively impacting the water quality. Aspiring to reach a more accurate and precise forecasting approach to predict levels of DO, the present work proposes new graph-based and transformer-based deep learning models. The models were trained and validated for the Credit River Watershed, and the results were compared with both benchmarking and literature-found approaches. The proposed Graph Neural Network Sample and Aggregate (GNN-SAGE) model was the best-performing approach, reaching coefficient of determination (R2) and Root Mean Squared Error (RMSE) values of 97% and 0.34 ppm, respectively. The findings from the Shapley additive explanations (SHAP) indicated that the GNN-SAGE benefited from spatiotemporal information from the surrounding stations, improving the model’s results, and that temperature is a major input attribute for determining future DO levels. The results established that the proposed GNN-SAGE model stands as a state-of-the-art solution for DO forecasting, with potential for real-time water quality applications.

In recent years, computer-aided analysis techniques have emerged as valuable tools in assisting dermatologists by providing objective and efficient analysis of skin cancer images. This paper utilizes the combination of the Contourlet Transform (CT) and Local Binary Pattern (LBP) techniques for accurately recognizing borders, contrast changes, and shapes of skin cancer images. These results often contain many features, leading to high computational costs and potential over-fitting issues. Hence, we applied Particle Swarm Optimization (PSO) to select the most informative and discriminating features, reducing the dimensionality while retaining important information for accurate classification. After reducing the feature set with PSO, we applied these sets to Machine learning classification algorithms: Support Vector Machine (SVM), Random Forest (RF), and Neural Networks (NN). The results show that SVM has the lowest time complexity of 0.0458 seconds, followed by the Neural Network at 0.08730 seconds, and the Random Forest model has the highest time complexity of 0.1622 seconds. The SVM and Neural Network models are faster to train than the Random Forest model, making them more suitable for real-time or latency-sensitive applications. We also compared our proposed model with the state-of-the-art models and obtained the accuracy of 86.9%, which is the highest among the models.

When dealing with limited sample sizes or missing values in time series data, achieving high accuracy in forecasting becomes challenging. This challenge is particularly common in wind energy forecasting, where many wind farms lack complete data. In this paper, we employ the widely-used ARIMA (Autoregressive Integrated Moving Average) model for time series forecasting. To address the difficulties posed by limited sample sizes and missing values, we adopt a unique approach by utilizing multiple ARIMA models for different datasets. A key distinction from other multiple models methods for time series forecasting lies in each model within our approach using its own dataset, making the fusion process more intricate. To overcome this, we incorporate neural networks and two techniques for model fusion: direct fusion with a learning mechanism and compensation for missing data through training. For each ARIMA model, we introduce a noise estimation method to circumvent the white noise assumptions inherent in ARIMA models. Finally, we apply these methods to wind energy prediction, and our experimental results showcase a significant enhancement in forecasting accuracy.

Facial Emotion Recognition (FER) is crucial across psychology, neuroscience, computer vision, and machine learning due to the diversified and subjective nature of emotions, varying considerably across individuals, cultures, and contexts. This paper investigates the impact of histogram equalization, data augmentation and model optimization strategies on three well-known FER datasets using pre-trained VGG models. Additionally, this paper showcases the effectiveness of different regularization techniques, callbacks, and learning schedulers in enhancing model performance by conducting extensive experiments. The model evaluation is discussed in terms of the following metrics: Accuracy, Area Under the Receiver Operating Characteristic Curve (AUC-ROC), Area Under the Precision-Recall Curve (AUC-PRC), and Weighted F1 score. Notably, the fine-tuned VGG models exhibit state-of-the-art performance compared to more complex conventional transfer learning models resulting in accuracies of 100%, 95.92%, and 69.65% on the CK+, KDEF, and FER2013 datasets, respectively.

Eutrophication is a major environmental issue with many negative consequences, such as hypoxia and harmful cyanotoxins production. Monitoring coastal eutrophication is a crucial, especially for island countries like the Republic of Cyprus, which are economically dependent on the touristic sector. Additionally, the open-sea aquaculture industry in Cyprus has been exhibiting an increase in the last decades and environmental monitoring to identify possible signs of eutrophication is mandatory according to the legislation. Therefore, in this modelling study, two different types of Artificial Neural Networks (ANNs) are developed based on in situ-data collected from stations located in the coastal waters of Cyprus. Theses ANNs aim to model the eutrophication phenomenon based on two different data-driven modelling procedures. Firstly, the self-organizing map (SOM) ANN examines several water quality parameters (specifically water temperature, salinity, nitrogen species, ortho-phosphates, dissolved oxygen and electrical conductivity) interactions with the Chlorophyll-a parameter. The SOM model enables us to visualize the monitored parameters relationships and to comprehend complex biological mechanisms related to Chlorophyll-a production. A second feed-forward ANN model is also developed for predicting the Chlorophyll-a levels. Based on this ANN model, several scenarios associated to the eutrophication-related water quality parameters can be extracted. The combination of these two ANNs models is considered a holistic modelling approximation for the identification of eutrophication scenarios, since it enables not only the prediction of the Chlorophyll-a parameter levels, but also the “capturing” of hidden biological mechanisms associated with algal production.

Arctic sea ice prediction holds significant importance for facilitating Arctic route planning, optimizing fisheries management, and advancing the field of sea ice dynamics research. While various deep learning models have been developed for sea ice prediction, they predominantly operate at the seasonal or sub-seasonal scale, often focusing on localized areas, and few cater to full-region daily scale prediction. This study introduces the use of spatiotemporal sequence data prediction models, namely, the convolutional LSTM (ConvLSTM) and predictive recurrent neural network (PredRNN), for the prediction of sea ice concentration (SIC). Our analysis reveals that, when solely utilizing SIC historical data as the input, the ConvLSTM model outperforms the PredRNN model in SIC prediction. To enhance the model's capacity to capture spatiotemporal relationships between multiple variables, we expanded the range of input data types to form the ConvLSTM-multi and PredRNN-multi models. Experimental findings demonstrate that the ConvLSTM-multi model excels in assimilating the influence of reanalysis data on sea ice within the sea ice edge region, thus exhibiting superior performance in predicting daily Arctic SIC over the subsequent 10 days. Furthermore, sensitivity tests on various model parameters highlight the substantial impact of sea surface temperature and prediction date on the accuracy of daily sea ice prediction. Additionally, meteorological and oceanographic parameters primarily affect the prediction accuracy of the thin ice region at the edge of the sea ice.

Ultrasound (US) is commonly used for the diagnosis of liver masses. Ensemble learning has been widely used for image classification, but its methods have not been fully optimized. This study was performed to investigate the usefulness of ensemble learning and compare a number of ensemble learning techniques using multiple convolutional neural network (CNN)-trained models for image classification of liver masses in US images. The US imaging data set was classified into four categories: benign liver tumor (BLT, 6320 images), liver cyst (LCY, 2320 images), metastatic liver cancer (MLC, 9720 images), and primary liver cancer (PLC, 7840 images). In this study, 250 test images were randomly selected for each class, for a total of 1000 images, and the remaining images were used for training. Sixteen different CNNs were used for to train and test the US images. All four types of ensemble learning—soft voting (SV), stacking (ST), weighted average voting (WAV), and weighted hard voting (WHV)—showed greater accuracy than the single CNN. All four types also showed significantly better deep learning performance than ResNeXt101 alone. For image classification of liver masses using US images, ensemble learning improved the performance of deep learning over a single CNN.

Variations in groundwater levels that are noticed in different parts of the world are typically the primary focus of research carried out in hydrology. The fluctuating levels of groundwater can be ascribed to several different factors, some of which include an increase in the demand for water, inappropriate irrigation practices, improper management of soil, and unregulated extraction from aquifers. It is necessary to effectively manage groundwater resources to have a trustworthy method of measuring and forecasting groundwater levels. The dynamics of groundwater are inherently uneven and difficult to understand. As a result, applying methods driven by data could potentially yield significant gains in hydrology. In this study, two data-driven models were utilized to estimate groundwater levels at a total of 4 monitoring wells located in Colorado state, USA. These models included ARIMA and ANN. The comparative analysis of the algorithms made use of a data set that had one month's worth of information from each of the years 1980 to 2019. The models were put through their paces, and the results of those tests were statistically and visually assessed. To evaluate the relative accuracy and precision of the models, the mean absolute error (MAE), mean square error (MSE), and root-mean-square error (RMSE) were used. When compared to ARIMA, the analysis demonstrates that ANN can produce the most accurate forecasts of groundwater levels in Colorado state USA.

Functional Connectivity analysis using Electroencephalography signals is common. The EEG signals are converted to networks by transforming the signals into a correlation matrix and analyzing the resulting networks. Here, four learning models, namely, Logistic Regression, Random Forest, Support Vector Machine, and Recurrent Neural Networks, are implemented on the correlation matrix data to classify them either on their psychometric assessment or the effect of therapy; The EEG data is trail-based/event-related. The classifications based on RNN provided higher accuracy( 74-88%) than the other three models( 50-78%). Instead of using individual graph features, a correlation matrix provides an initial test of the data. When compared with the time-resolved correlation matrix, it offered a 4-5% higher accuracy. The time-resolved correlation matrix is better suited for dynamic studies here; it provides lower accuracy when compared to the correlation matrix, a static feature.

Deep learning has been applied in many areas that have had a significant impact on applications that improves real-life challenges. The success of deep learning in a wide range of areas is due in part to the use of activation functions, which are particularly effective at solving non-linear problems. Activation functions are a key focus for researchers in artificial intelligence who aim to improve the performance of neural networks. This article provides a comprehensive explanation and comparison of different activation functions, with a focus on the arc tangent and its variations specifically. The paper presents experimental results that show that variations of the arc tangent using irrational numbers such as pi, the golden ratio, and Euler’s number, as well as a self-arctan function, produce promising results. Since we experimented with promising activation functions on two different problems, and datasets, we reached a result that different irrationals work well for different problems. In other words, arctan ϕ gives the best results mostly for multiclass classification and arctan e gives the best results for time series prediction problems. The paper focuses on a multiclass classification problem applied to the Reuters Newswire dataset and a time-series prediction problem on Türkiye energy trade value to show the impacts of activation functions.

The integration of deep learning methods, especially Convolutional Neural Networks (CNN), and Synthetic Aperture Radar Automatic Target Recognition (SAR ATR) has been widely deployed in the field of radar signal processing. Nevertheless, these methods are frequently regarded as black-box models due to the limited visual interpretation of their internal feature representation and parameter organization. In this paper, we propose an innovative approach named C-RISE, which builds upon the RISE algorithm to provide a post-hoc interpretation technique for black-box models used in SAR Images Target Recognition. C-RISE generates saliency maps that effectively visualize the significance of each pixel. Our algorithm outperforms RISE by clustering masks that capture similar fusion features into distinct groups, enabling more appropriate weight distribution and increased focus on the target area. Furthermore, we employ Gaussian blur to process the masked area, preserving the original image structure with optimal consistency and integrity. C-RISE has been extensively evaluated through experiments, and the results demonstrate superior performance over other interpretation methods based on perturbation when applied to neural networks for SAR image target recognition. Furthermore, our approach is highly robust and transferable compared to other interpretable algorithms, including white-box methods.

Deep learning has been applied in many areas that have had a significant impact on applications that improves real-life challenges. The success of deep learning in a wide range of areas is due in part to the use of activation functions, which are particularly effective at solving non-linear problems. Activation functions are a key focus for researchers in artificial intelligence who aim to improve the performance of neural networks. This article provides a comprehensive explanation and comparison of different activation functions, with a focus on the arc tangent and its variations specifically. The paper presents experimental results that show that variations of the arc tangent using irrational numbers such as pi, the golden ratio, and Euler’s number, as well as a self-arctan function, produce promising results. Since we experimented with promising activation functions on two different problems, and datasets, we reached a result that different irrationals work well for different problems. In other words, arctan ϕ gives the best results mostly for multiclass classification and arctan e gives the best results for time series prediction problems. The paper focuses on a multi-class classification problem applied to the Reuters Newswire dataset and a time-series prediction problem on Türkiye energy trade value to show the impacts of activation functions.

Total cloud cover (TCC) retrieval from ground-based optical imagery is a problem being tackled by a few generations of researchers. The number of human-designed algorithms for the estimation of TCC grows every year. However, there is not very much progress in terms of quality, mostly due to the lack of systematic approach to the design of the algorithms, to the assessment of their generalization ability, and to the assessment of the TCC retrieval quality. In this study, we discuss the optimization nature of data-driven schemes for TCC retrieval. In order to compare the algorithms, we propose the framework for the assessment of the algorithms characteristics. We present several new algorithms that are based on deep learning techniques: a model for outliers filtering, and a few models for TCC retrieval from all-sky imagery. For training and assessment of data-driven algorithms of this study, we present the Dataset of All-Sky Imagery over the Ocean (DASIO) containing over one million of all-sky optical images of visible sky dome taken in various regions of the World Ocean. The research campaigns contributed to DASIO collection took place in the Atlantic ocean, the Indian ocean the Red and Mediterranean seas, and also in the Arctic ocean. Optical imagery collected during these missions are accompanied by standard meteorological observations of cloudiness characteristics made by experienced observers. We assess the generalization ability of the presented models in several scenarios that differ in terms of the regions selected for the train and validation subsets. As a result, we demonstrate that our models based on convolutional neural networks deliver superior quality compared to all previously published schemes. As a key result, we demonstrate considerable drop of the ability to generalize the training data in case of strong covariate shift between training and validation subsets of imagery which may take place in case of region-aware subsampling.

Substantial evidence indicates that bisphenol A (BPA), a ubiquitous environmental chemical used in the synthesis of polycarbonate plastics and epoxy resins, can impair brain development. Clinical and epidemiological studies exploring potential connections between BPA and neurodevelopmental disorders in humans have repeatedly identified correlations between early BPA exposure and developmental disorders, like attention deficit/hyperactivity disorder and autism spectrum disorder. Investigations using invertebrate and vertebrate animal models have revealed that developmental exposure to BPA can impair multiple aspects of neuronal development, including neural stem cell proliferation and differentiation, synapse formation, and synaptic plasticity—neuronal phenotypes that are thought to underpin the fundamental changes in behavior associated neurodevelopmental disorders. Consistent with BPA-associated neuronal phenotypes, behavioral analyses of BPA-treated animals have shown significant impacts on behavioral endophenotypes related to neurodevelopmental disorders, including altered locomotor activity, learning and memory deficits, and anxiety-like behavior. To contextualize the correlations between BPA and neurodevelopmental disorders in humans, this review summarizes current literature reporting on the developmental neurotoxicity of BPA in laboratory animals, with an emphasis on neuronal phenotypes, molecular mechanisms, and behavioral outcomes. The collective works described here predominantly support the notion that gestational exposure to BPA should be regarded as a risk factor for neurodevelopmental disorders.

Indian cities are increasingly becoming susceptible to PM10 induced health effects which have become a matter of concern for the policymakers of the country. Air pollution is engulfing the comparatively smaller cities as the rapid pace of urbanization, and economic development seems not to lose steam ever. A review of air pollution of 28 cities of India, which includes tier-I, II, and III cities of India, found to have grossly violated both WHO and NAAQS standards in respect of acceptable daily average PM10 concentrations by a wide margin. Predicting the city level PM10 concentrations in advance and accordingly initiate prior actions is an acceptable solution to save the city dwellers from PM10 induced health hazards. Predictive ability of three models, linear MLR, nonlinear MLP (ANN), and nonlinear CART, for one day ahead PM10 concentration forecasting of tier-II Guwahati city, were tested with 2016-2018 daily average observed climate data, PM10, and gaseous pollutants. The results show that the non-linear algorithm MLP with feedforward backpropagation network topologies of ANN class, giving the best prediction value when compared with linear MLR and nonlinear CART model. ANN (MLP) approach, therefore, may be useful to effectively derive a predictive understanding of one day ahead PM10 concentration level and thus provide a tool to the policymakers for improving decision-making associated with air pollution and public health.

Waste heat recovery stands out as a promising technique to tackle both energy shortages and environmental pollution. Currently, this valuable resource, generated from processes like fuel combustion or chemical reactions, is often dissipated into the environment, despite its potential to significantly contribute to the economy. To harness this untapped potential, a travelling-wave thermo-acoustic generator has been designed and subjected to comprehensive experimental analysis. Fifty-two data corresponding to different working conditions of the system were extracted to build ANN, ANFIS and ANN-PSO models. Evaluation of performance metrics reveals that the ANN-PSO model demonstrates the highest predictive accuracy (R^2=0.9959), particularly in relation to output voltage. This research demonstrates the potential of machine learning techniques for the analysis of thermo-acoustic systems. By doing so, it is possible to get an insight into nonlinearities inherent to thermo-acoustic systems. This advancement empowers researchers to forecast the performance characteristics of alternative configurations with a heightened level of precision.

Scoring targets in shooting sports is a crucial and time-consuming task that relies on manually counting bullet holes. This paper introduced an automatic score detection model using object detection techniques. The performance of seven models belonging to two different architectural setups was compared. Models like YOLOv8n, YOLOv8s, YOLOv8m, RetinaNet-50, and RetinaNet-101 are single-shot detectors, while Faster RCNN-50 and Faster RCNN-101 belong to the two-shot detectors category. The dataset was manually captured from the shooting range and expanded by generating more versatile data using Python code. Before the dataset was trained to develop models, it was resized (640x640) and augmented using Roboflow API. The trained models were then assessed on the test dataset, and their performance was compared using matrices like mAP50, mAP50-90, precision, and recall. The results showed that YOLOv8 models can detect multiple objects with good confidence scores. Among all, YOLOv8m performed the best with the highest mAP50 value of 96.7%, followed by the performance of YOLOv8s with the mAP50 value of 96.5%. It is suggested that if the system is to be implemented in a real-time environment, YOLOv8s is a better choice since it took significantly less inference time (2.3ms) than YOLOv8m (5.7ms) and yet generated the competitive mAP50.

This article explores the crucial task of estimating State of Charge (SOC) in Hybrid Electric Vehicles (HEVs) and examines the applicability of various regression models for this purpose. We delve into the strengths and limitations of Linear Regression, Support Vector Regression (SVR), and Neural Network Regression (NNR) in the context of SOC estimation. Linear regression provides a simple and interpretable baseline, SVR extends this to nonlinear relationships, while NNR emerges as a powerful tool with adaptive capabilities. The choice of model depends on factors such as data characteristics, interpretability, and computational resources. As the field evolves, the article advocates for a nuanced approach, possibly incorporating hybrid models, to achieve robust and accurate SOC estimation, contributing to the ongoing enhancement of HEV efficiency and sustainability.

High-rise building machines (HBMs) play a critical role in the successful construction of super-high skyscrapers, providing essential support and ensuring safety. The HBM’s climbing system relies on a jacking mechanism consisting of several independent jacking cylinders. A reliable control system is imperative to maintain a smooth posture of the construction steel platform (SP) under the action of the jacking mechanism. This research introduces three multivariate time series neural network models—namely, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Temporal Convolutional Network (TCN)—to predict the posture of HBM. The models take pressure and stroke measurements from the jacking cylinders as inputs, and their outputs determine the levelness of the SP and the posture of the HBM at various climbing stages. The development and training of these neural networks are based on historical on-site data, with the predictions subjected to thorough comparative analysis. All proposed neural network models exhibit the capability to dynamically predict the posture of the HBM during the climbing process, using data from sensors. Notably, the GRU model shows better predictive performance.

The variability of the material properties of steel from different suppliers causes problems in achieving the required surface quality after turning. Therefore, the manufacturer needs to be able to estimate the resulting quality before starting production, especially if it is an expensive, small-batch production from stainless steel. Predictive models will make it possible to estimate the surface roughness from the mechanical properties of steel in the Mill Test Certificate (MTC) and thus support decision-making about the supplier selection or the acceptance of a material supply. Multivariate second-order polynomial model and feedforward back-propagation Artificial Neural Network (ANN) model were used to enhance the methodological robustness in formulating the decision if the predicted surface roughness is outside the required range, even before accepting the delivery. Both models can accurately predict surface roughness, while the ANN model is more accurate than the polynomial model; however, the predictive model is sensitive to the accuracy of the input data, and the model’s prediction is valid only under precisely defined conditions. The prediction model is used in a step-by-step decision-making procedure, which enables the trained staff to make a quick decision.

Electricity demand prediction is vital for energy production management and proper exploitation of the present resources. Recently, several novel machine learning (ML) models have been employed for electricity demand prediction to estimate the future prospects of the energy requirements. The main objective of this study is to review the various ML models applied for electricity demand prediction. Through a novel search and taxonomy, the most relevant original research articles in the field are identified and further classified according to the ML modeling technique, perdition type, and the application area. A comprehensive review of the literature identifies the major ML models, their applications and a discussion on the evaluation of their performance. This paper further makes a discussion on the trend and the performance of the ML models. As the result, this research reports an outstanding rise in the accuracy, robustness, precision and the generalization ability of the prediction models using the hybrid and ensemble ML algorithms.

Machine learning (ML) has a large capacity to learn and analyze a large volume of data. This study aimed to train different algorithms to discriminate between healthy and pathologic corneal images by evaluating digitally processed spectral-domain optical coherence tomography (SD-OCT) corneal images. A set of 22 SD-OCT images belonging to a random set of corneal pathologies was compared to 71 healthy corneas (control group). A binary classification method was applied; three approaches of ML were used. Once all images were analyzed, representative areas from every digital image were also processed and analyzed for a statistical feature comparison between healthy and pathologic corneas. The best performance was obtained from transfer learning - support vector machine (TL-SVM) (AUROC = 0.94, SPE 88%, SEN 100%) and transfer learning – random forest (TL- RF) method (AUROC = 0.92, SPE 84%, SEN 100%), followed by convolutional neural network (CNN) (AUROC = 0.84, SPE 77%, SEN 91%) and random forest (AUROC = 0.77, SPE 60%, SEN 95%). The highest diagnostic accuracy in classifying corneal images was achieved with the TL-SVM and the TL-RF models. In image classification, CNN was a strong predictor. This pilot experimental study developed a systematic mechanized system to discern pathologic from healthy corneas.

In recent years, the unmanned aerial vehicle (UAV) synthetic aperture radar (SAR) has become a highly sought-after topic for its wide applications in the field of target recognition, detection, and tracking. However, SAR automatic target recognition (ATR) models based on deep neural networks (DNN) are suffering from adversarial examples. Generally, non-cooperators rarely disclose any information about SAR-ATR models, making adversarial attacks challenging. In this situation, we propose Transferable Adversarial Network (TAN) to attack these models with highly transferable adversarial examples. The proposed method improves the transferability via a two-player game, in which we simultaneously train two encoder-decoder models: a generator that crafts malicious samples through a one-step forward mapping from original data, and an attenuator that weakens the effectiveness of malicious samples by capturing the most harmful deformations. In particular, compared to traditional iterative methods, our approach is able to one-step map original samples to adversarial examples, thus enabling real-time attacks. Experimental results indicate that the proposed approach achieves state-of-the-art transferability with acceptable adversarial perturbations and minimum time costs compared to existing attack methods, i.e., it excellently realizes real-time transferable adversarial attacks.

Byers Peninsula (62°34’-62°40’S-60°54’61°13’W), 60 km2 in size, is considered one of the largest ice-free areas in Antarctica. Since 2006, the Spanish Polar Program has taken part in a large number of environmental studies involving effects of climate changes, limnology and microbiology, live cycles, but not albedo. Surface albedo is one of the key physical parameters in the surface energy budget of polar regions. Most of Antarctica is covered by ice sheets; only about 0.44% of the area is permanently ice-free. However, in maritime Antarctica, the ice-free areas, corresponding to small islands, peninsulas, and coastal beaches, account for about 3% of the territory. To incorporate the contribution of these areas into global albedo models, it is necessary to relate the soil properties of the ice-free areas to the surface albedo response. Also, mapping soil properties and albedo have and special interest in these ice-free areas. Image classification using machine learning methods trained with georeferenced soil data could be useful for mapping soil properties and albedo in multispectral optical satellite images. A shallow neural network implemented using the Keras Python module was used to define and train models of soil properties using 15 explanatory variables corresponding to bands and spectral indices of a Sentinel image and a population of 49 soil samples taken from the top 5 centimeters of the soil profile. The soil samples were analyzed in the laboratory and a spectral library in the Vis-Nir range (350-2500 nm) was created. The albedo of the samples was integrated from the ADS spectra. At the same time, a linear regression model of albedo using the soil properties as explanatory variables was performed. The R2 fit of this new model was about 0.82 and the error of estimation was 4.1. The model was extended to the entire Byers Peninsula using the ML soil property models as explanatory variables. The RMSE of the extended models increased up to 8.2. We think that the main cause of this increase (2.6 points) is related to the error propagation, phenomenon due to the use of models as explanatory variables. However, another important part of the error increase could be due, among other reasons, to the use of an image that, although corrected, is not completely free of clouds.

Interferometry Synthetic Aperture Radar (InSAR) is an advanced remote sensing technique for studying the earth's surface topography and deformations. It is used to generate high-quality Digital Elevation Models (DEMs). DEMs are a crucial and primary input to various topographical quantification and modelling applications. The quality of input DEMs can be further improved using fusion methods, which combine multi-sensor or multi-temporal datasets intelligently to retrieve the best information amongst the input data. This research study is based on developing a Neural Network based fusion approach for improving InSAR based DEMs in plain and hilly terrains. The study areas comprise of relatively plain terrain from Ghaziabad and hilly terrain of Dehradun and their surrounding regions. The training dataset consists of DEM elevations and derived topographic attributes like slope, aspect, topographic position index (TPI), terrain ruggedness index (TRI), and vector roughness measure (VRM) in different land use land cover classes of the study areas. The spaceborne altimetry ICESat-2 ATL08 photon data is used as a reference elevation. A Feed Forward Neural Network with backpropagation algorithm is trained based on the prepared training samples. The trained model produces fused DEMs by learning the relationship between the input and target samples. This is used to predict elevations in the test areas. The accuracy of results from the models are assessed with TanDEM-X 90 m DEM. The fused DEMs show significant improvement in terms of RMSE over the input DEMs with improvement factor of 94.65 % in plain area and 82.62 % in hilly area. The study concludes that the ANN with its universal approximation property is able to significantly improve the fused DEM.

Keywords: Artificial Neural Network (ANN), Support Vector Machine (SVM), Support Vector Regression (SVR), Lightweight Gradient Boosting Machines (Light GBM), Machine Learning, Solar Irradiance (SI), Solar forecasting

Over the years, machine learning has emerged as a tool for automated translation and has been studied relentlessly for decades. RBMT, SMT, and NMT models have been used to achieve machine translation and the results have drastically improved from when research in this field first began. Although a few general-purpose translators such as Google Translate or Microsoft Translator have accurate translations compared to that of a human translator, many pieces of text containing highly technical terms or homonyms are often mistranslated completely. When considering the necessity and importance of translating technical patents from different domains, accuracy in translation is not something that can be compromised. This motivates the need to improve the performance of machine translation further. The scope of this paper covers three open-source machine translation models for the purpose of patent documentation translation from English to Japanese, evaluates their performance on patent data, and proposes a methodology that enabled us to improve one of the model’s BLEU score by 41.22%, achieving a BLEU score of 46.18.

Solar energy is a renewable resources of energy which is broadly utilized and have the least pollution impact between the available alternatives of fossil fuels. In this investigation, machine leaening approaches of neural networks (NN), neuro-fuzzy and least squares support vector machine (LSSVM) are used to build the models for prediction of the thermal performance of a photovoltaic-thermal solar collector (PV/T) by estimating its efficiency as an output of the model while inlet temperature, flow rate, heat, solar radiation, and heat of sun are input of the designed model. Experimental measurements was prepared by designing a solar collector system and 100 data extracted. Different analyses are also performed to examine the credibility of the introduced approaches revealing great performance. The suggested LSSVM model represented the best performance regarding the mean squared error (MSE) of 0.003 and correlation coefficient (R²) value of 0.99, respectively.

Prediction models in mobility and transportation maintenance systems have been dramatically improved through using machine learning methods. This paper proposes novel machine learning models for an intelligent road inspection. The traditional road inspection systems based on the pavement condition index (PCI) are often associated with the critical safety, energy and cost issues. Alternatively, the proposed models utilize surface deflection data from falling weight deflectometer (FWD) tests to predict the PCI. Machine learning methods are the single multi-layer perceptron (MLP) and radial basis function (RBF) neural networks as well their hybrids, i.e., Levenberg-Marquardt (MLP-LM), scaled conjugate gradient (MLP-SCG), imperialist competitive (RBF-ICA), and genetic algorithms (RBF-GA). Furthermore, the committee machine intelligent systems (CMIS) method was adopted to combine the results and improve the accuracy of the modeling. The results of the analysis have been verified through using four criteria of average percent relative error (APRE), average absolute percent relative error (AAPRE), root mean square error (RMSE), and standard error (SD). The CMIS model outperforms other models with the promising results of APRE=2.3303, AAPRE=11.6768, RMSE=12.0056, and SD=0.0210.

Downward short-wave (SW) solar radiation is the only essential energy source powering the atmospheric dynamics, ocean dynamics, biochemical processes, etc. on our planet. Clouds are the main factor limiting the SW flux over the land and the Ocean. For the accurate meteorological measurements of the SW flux one needs expensive equipment - pyranometers. For some cases where one does not need golden-standard quality of measurements, we propose estimating incoming SW radiation flux using all-sky optical RGB imagery which is assumed to incapsulate the whole information about the SW flux. We used DASIO all-sky imagery dataset with corresponding SW downward radiation flux measurements registered by an accurate pyranometer. The dataset has been collected in various regions of the World Ocean during several marine campaigns from 2014 to 2022, and it will be updated. We demonstrate the capabilities of several machine learning models in this problem, which are multilinear regression, Random Forests, Gradient Boosting and convolutional neural networks (CNN). We also applied the inverse target frequency (ITF) re-weighting of the training subset and the approach of pre-classification in order to raise the SW flux approximation quality. We found that the CNN is capable of approximating downward SW solar radiation with higher accuracy compared to existing empiric parameterizations. We also found that the ITF reweighting of the training subset increases the approximation quality substantially. The estimates of the SW radiation flux using all-sky imagery may be of particular use in case of the need for the fast radiative budgets assessment of a site.

Oil spillage over a sea or ocean’s surface is a threat to marine and coastal ecosystems. Spaceborne synthetic aperture radar (SAR) data has been used efficiently for the detection of oil spills due to its operational capability in all-day all-weather conditions. The problem is often modeled as a semantic segmentation task. The images need to be segmented into multiple regions of interest such as sea surface, oil spill, look-alikes, ships and land. Training of a classifier for this task is particularly challenging since there is an inherent class imbalance. In this work, we train a convolutional neural network (CNN) with multiple feature extractors for pixel-wise classification; and introduce to use a new loss function, namely ‘gradient profile’ (GP) loss, which is in fact the constituent of the more generic Spatial Profile loss proposed for image translation problems. For the purpose of training, testing and performance evaluation, we use a publicly available dataset with selected oil spill events verified by the European Maritime Safety Agency (EMSA). The results obtained show that the proposed CNN trained with a combination of GP, Jaccard and focal loss functions can detect oil spills with an intersection over union (IoU) value of 63.95%. The IoU value for sea surface, look-alikes, ships and land class is 96.00%, 60.87%, 74.61% and 96.80%, respectively. The mean intersection over union (mIoU) value for all the classes is 78.45%, which accounts for a 13% improvement over the state of the art for this dataset. Moreover, we provide extensive ablation on different Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) based hybrid models to demonstrate the effectiveness of adding GP loss as an additional loss function for training. Results show that GP loss significantly improves the mIoU and F1 scores for CNNs as well as ViTs based hybrid models. GP loss turns out to be a promising loss function in the context of deep learning with SAR images.

In the present work, a novel and the robust computational investigation is carried out to estimate solubility of different acids in supercritical carbon dioxide. Four different algorithms such as radial basis function artificial neural network, Multi-layer Perceptron (MLP) artificial neural network (ANN), Least squares support vector machine (LSSVM) and adaptive neuro-fuzzy inference system (ANFIS) are developed to predict the solubility of different acids in carbon dioxide based on the temperature, pressure, hydrogen number, carbon number, molecular weight, and acid dissociation constant of acid. In the purpose of best evaluation of proposed models, different graphical and statistical analyses and also a novel sensitivity analysis are carried out. The present study proposed the great manners for best acid solubility estimation in supercritical carbon dioxide, which can be helpful for engineers and chemists to predict operational conditions in industries.

One of the most effective image processing techniques is the use of convolutional neural networks, where we combine intensity values at grid points in the vicinity of each point. To speed up computations, researchers have developed a dilated version of this technique, in which only some points are processed. It turns out that the most efficient case is when we select points from a sub-grid. In this paper, we explain this empirical efficiency proving that the sub-grid is indeed optimal – in some reasonable sense. To be more precise, we prove that all reasonable optimality criteria, the optimal subset of the original grid is either a sub-grid, or a sub-grid-like set.

Optical artificial neural networks (ONNs) have significant potential for ultra-high computing speed and energy efficiency. We report a new approach to ONNs based on integrated Kerr micro-combs that is programmable, highly scalable and capable of reaching ultra-high speeds, demonstrating the building block of the ONN — a single neuron perceptron — by mapping synapses onto 49 wavelengths to achieve a single-unit throughput of 11.9 Giga-OPS at 8 bits per OP, or 95.2 Gbps. We test the perceptron on handwritten-digit recognition and cancer-cell detection — achieving over 90% and 85% accuracy, respectively. By scaling the perceptron to a deep learning network using off-the-shelf telecom technology we can achieve high throughput operation for matrix multiplication for real-time massive data processing.

The informational synthesis of neural structures, processes, parameters and characteristics that allow a unified description and modeling as neural machines of natural and artificial neural systems is presented. The general informational parameters as the global quantitative measure of the neural systems computing potential as absolute and relative neural power were proposed. Neural information organizing and processing follows the way in which nature manages neural information by developing functions, functionalities and circuits related to different internal or peripheral components and also to the whole system through a non-deterministic memorization, fragmentation and aggregation of afferent and efferent information, deep neural information processing representing multiple alternations of fragmentation and aggregation stages. The relevant neural characteristics were integrated into a neural machine type model that incorporates unitary also peripheral or interface components as the central ones. The proposed approach allows overcoming the technical constraints in artificial computational implementations of neural information processes and also provides a more relevant description of natural ones.

The intracranial pressure (ICP) signal, as monitored on patients in intensive care units, contains pulses of cardiac origin where P1 and P2 subpeaks can often be observed. When calculable, the ratio of their relative amplitudes is an indicator of the patient’s cerebral compliance. This characterization is particularly informative on the overall state of the cerebrospinal system. We developed a recurrent neural network-based framework for P2/P1 ratio computation that only takes a raw ICP signal as an input. Two tasks are performed, namely pulse classification and subpeak designation. Pulse classification was achieved with an area under the curve of 0.90 on a 4,344-pulse testing dataset, while the peak designation algorithm identified pulses with a P2/P1 ratio > 1 with a 97.3% accuracy. Although it still needs to be evaluated on a larger number of labeled recordings, our automated P2/P1 ratio calculation framework appears to be a promising tool for improving bedside monitoring devices.

An earthquake is a natural event by its general definition. This natural event is a disaster that causes significant damage, loss of life, and other economic effects that will damage the state. The possibility of predicting a natural event such as an earthquake will minimize the reasons mentioned. Data collection, data processing, and data evaluation were carried out in this study. Earthquake forecasting was performed using the data and the RNN (Recurrent Neural Network) method. The study was carried out on seismic data with a magnitude of 3.0 and above belonging to Düzce Province between 1990 and 2022. In order to increase the learning potential of the method, the b and d values of the earthquake were calculated and included in the data set, except for the earthquake magnitude. The determination of earthquakes in a specific time interval in regions of Turkey, the classification of earthquake-related seismic data using artificial neural networks, and the production of predictions for the future reveal the importance of this study.

This paper investigates the dynamic properties of artificial neural networks using differential equations and explores the influence of parameters on stability and neural oscillations. By analyzing the equilibrium point of the differential equations, we identify conditions for asymptotic stability and criteria for oscillation in artificial neural networks. Furthermore, we demonstrate how adjusting synaptic weights between neurons can effectively control stability and oscillation. The proposed model offers potential insights into the malfunctioning mechanisms of biological neural networks implicated in neurological disorders like Parkinson's disease tremors and epilepsy seizures, which are characterized by abnormal oscillations.

Multiply-accumulate calculations using a memristor crossbar array is an important method to realize neuromorphic computing. However, the memristor array fabrication technology is still immature, and it is difficult to fabricate large-scale arrays with high-yield, which restricts the development of memristor-based neuromorphic computing technology. Therefore, cascading small-scale arrays to achieve the neuromorphic computational ability that can be achieved by large-scale arrays, which is of great significance for promoting the application of memristor-based neuromorphic computing. To address this issue, we present a memristor-based cascaded framework with some basic computation units, several neural network processing units can be cascaded by this means to improve the processing capability of the dataset. Besides, we introduce a split method to reduce pressure of input terminal. Compared with VGGNet and GoogLeNet, the proposed cascaded framework can achieve 93.54% Fashion-MNIST accuracy under the 4.15M parameters. Extensive experiments with Ti/AlOx/TaOx/Pt we fabricated are conducted to show that the circuit simulation results can still provide a high recognition accuracy, and the recognition accuracy loss after circuit simulation can be controlled at around 0.26%.

Tumorigenesis is a complex biological phenomenon that includes extensive genetic and phenotypic heterogeneities and complicated regulatory mechanisms. In the recent few years, our studies demonstrate that tumor-initiating cells are similar to neural stem/progenitor cells in regulatory networks, tumorigenicity and pluripotent differentiation potential. In the review, I will make further discussion on these observations and propose a rule of cell biology by integrating these findings with evidence from developmental biology, tumor biology and evolution, which suggests that neural stemness underlies two coupled cell properties, tumorigenicity and pluripotent differentiation potential. Tumorigenicity and phenotypic heterogeneity in tumor is a result of acquirement of neural stemness in cells. The neural stemness property of tumor-initiating cells can hopefully integrate different concepts/hypotheses underlying tumorigenesis. Neural stem cells/neural progenitors and tumor-initiating cells share regulatory networks; both exhibit neural stemness, tumorigenicity and differentiation potential; both are dependent on expression or activation of ancestral genes (the atavistic effect); both rely primarily on aerobic glycolytic metabolism; both can differentiate into various cells or tissues that are derived from three germ layers, resembling severely disorganized or more severely degenerated process of embryonic development; both are enriched in long genes with more splice variants that provide more plastic scaffolds for cell differentiation, etc. The property of neural stemness might be a key point to understand tumorigenesis and pluripotent differentiation potential, and possibly explain certain pathological observations in tumors that have been inexplicable. Therefore, behind the complexity of tumorigenesis might be a general rule of cell biology, i.e., neural stemness represents the ground state of cell tumorigenicity and pluripotent differentiation potential.

This research investigates the application of deep learning in sentiment analysis of Canadian maritime case law. It offers a framework for improving maritime law and legal analytics policy-making procedures. The automation of legal document extraction takes center stage, underscoring the vital role sentiment analysis plays at the document level. Therefore, this study introduces a novel strategy for sentiment analysis in Canadian maritime case law, combining sentiment case law approaches with state-of-the-art deep learning techniques. The overarching goal is to systematically unearth hidden biases within case law and investigate their impact on legal outcomes. Employing Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) based models, this research achieves a remarkable accuracy of 98.05% for categorizing instances. In contrast, Conventional Machine Learning techniques such as Support Vector Machines (SVM) yield an accuracy rate of 52.57%, Naive Bayes at 57.44 %, and Logistic Regression at 61.86%. The superior accuracy of the CNN and STM model combination underscores its usefulness in legal sentiment analysis, offering promising future applications in diverse fields like legal analytics and policy design. These findings mark a significant choice for AI-powered legal tools, presenting more sophisticated and sentiment-aware options for the legal profession.

A chronically stable electrode material with low impedance for recording neural activity and high charge-injection capacity for functional electro-stimulation is desirable for the fabrication of implantable microelectrode arrays that aims to restore impaired or lost neurological functions in humans. For this purpose, we have investigated the electrochemical properties of sputtered ruthenium oxide (RuOx) electrode coatings, deposited on planar microelectrode arrays, using an inorganic model of interstitial fluid (model-ISF) at 37o C as the electrolyte. Through a combination of cyclic voltammetry (CV) and electrochemical impedance spectroscopy (EIS) modelling study we have established the contribution of faradaic reaction as the major charge-injection contributor within the safe neural stimulation potential window of 0.6 V vs Ag|AgCl. We have also established the reversibility of the charge-injection process for sputtered RuOx film by applying constant charge-per-phase current stimulations at different pulse widths and by comparing the magnitudes of leading and trailing access voltages during voltage transient measurements. Finally, the impedance of the sputtered RuOx film was found to be reasonably comparable in both its oxidized and reduced states, although electronic contribution from capacitive double-layer was found to be slightly higher for the completely oxidized film around 0.6 V than its reduced counterpart around -0.6V.

Convolutional neural networks (CNNs), inspired by biological visual cortex systems, are a powerful category of artificial neural networks that can extract the hierarchical features of raw data to greatly reduce the network parametric complexity and enhance the predicting accuracy. They are of significant interest for machine learning tasks such as computer vision, speech recognition, playing board games and medical diagnosis [1-7]. Optical neural networks offer the promise of dramatically accelerating computing speed to overcome the inherent bandwidth bottleneck of electronics. Here, we demonstrate a universal optical vector convolutional accelerator operating beyond 10 Tera-OPS (TOPS - operations per second), generating convolutions of images of 250,000 pixels with 8-bit resolution for 10 kernels simultaneously — enough for facial image recognition. We then use the same hardware to sequentially form a deep optical CNN with ten output neurons, achieving successful recognition of full 10 digits with 900 pixel handwritten digit images with 88% accuracy. Our results are based on simultaneously interleaving temporal, wavelength and spatial dimensions enabled by an integrated microcomb source. This approach is scalable and trainable to much more complex networks for demanding applications such as unmanned vehicle and real-time video recognition.

This research investigates the implementation of complex exponential-based neurons in FPGA, which can pave the way for implementing bio-inspired spiking neural networks to compensate for the existing computational constraints in conventional artificial neural networks. The increasing use of extensive neural networks and the complexity of models in handling big data lead to higher power consumption and delays. Hence, finding solutions to reduce computational complexity is crucial for addressing power consumption challenges. The complex exponential form effectively encodes oscillating features like frequency, amplitude, and phase shift, streamlining the demanding calculations typical of conventional artificial neurons through levering simple phase addition of complex exponential functions. The article implements such a two-neuron and a multi-neuron neural model using Xilinx system generator and Vivado design suite, employing 8-bit, 16-bit, and 32-bit fixed-point data format representations. The study evaluates the accuracy of the proposed neuron model across different FPGA implementations while also providing a detailed analysis of operating frequency, power consumption, and resource usage for the hardware implementations. BRAM-based Vivado designs outperformed Simulink regarding speed, power, and resource efficiency. Specifically, the Vivado BRAM-based approach supported up to 128 neurons, showcasing optimal LUT and FF resource utilization. Such outcomes accommodate choosing the optimal design procedure for implementing spiking neural networks on FPGAs.

Neural stem cell (NSC) based therapies are at the forefront of regenerative medicine strategies to combat illness and injury of the central nervous system (CNS). In addition to their ability to produce new cells, NSCs secrete a variety of products, known as the NSC secretome, that have been shown to ameliorate CNS disease pathology and promote recovery. As pre-clinical and clinical research to harness the NSC secretome for therapeutic purposes advances, a more thorough understanding of the endogenous NSC secretome can provide useful insight into the functional capabilities of NSCs. In this review, we focus on research investigating the autocrine and paracrine functions of the endogenous NSC secretome across life. We also compare the NSC secretome across species, finding signs of conserved parallels between rodent, human and zebrafish NSC secretomes. Throughout development and adulthood, we find evidence that the NSC secretome is a critical component of how endogenous NSCs regulate themselves and their niche. We also find gaps in current literature, most notably in the clinically relevant domain of endogenous NSC paracrine function in the injured CNS. Future investigations to further define the endogenous NSC secretome and its role in CNS tissue regulation are necessary to bolster our understanding of NSC-niche interactions and to aid in the generation of safe and effective NSC-based therapies.

Biological neural network (BNN) is the core brain for creature to accomplish its intelligent behaviors through unique network structure and mechanisms. It also inspires controlling human-designed autonomous agents including robots to generate more advanced intelligent behaviors by mimicking the neural control mechanism of creatures at a deeper and more interactive level. Here, we constructed a whole-brain neural network model of Caenorhabditis elegans (C. elegans), which characterizes the electrochemical processes at the level of cellular synapse. The neural network simulation integrates computational programming and visualization of neurons and synapse connections of C. elegans, containing the specific controllable circuits summarization with their dynamic characteristics within the whole network. To illustrate its particular intelligent control capability in terms of robotics, we introduce the first innovative methodology for applying the BNN model to a 12-legged robot’s movement control based on the established numerical simulation platforms. We accomplished and designed two methods and corresponding encoding processes, one involving orientation control and the other involving locomotion generation, to demonstrate the intelligent control performance of BNN. Both simulation and experiment results indicate that the robot exhibits stronger autonomy and more intelligent moving performance under the control of BNN. We then summarized the contributions for digitalizing the C. elegans’ whole-brain neural network in real-time and utilizing it to control robot in a closed loop to validate its advanced intelligence control ability in terms of a scientific robot.

Speech recognition, or automatic speech recognition (ASR), is a technology designed to convert spoken language into text using software. However, conventional ASR methods involve several distinct components, including language, acoustic, and pronunciation models with dictionaries. This modular approach can be time-consuming and may influence performance. In this study, we propose a method that streamlines the speech recognition process by incorporating a unified recurrent neural network (RNN) architecture. Our architecture integrates a convolutional neural network (CNN) with an RNN and employs a connectionist temporal classification (CTC) loss function. Key experiments were carried out using a dataset comprising 576,656 valid sentences, using erosion techniques. Evaluation of the model performance, measured by the word error rate (WER) metric, demonstrated remarkable results, achieving a WER of 2%. This approach has significant implications for the realm of speech recognition, as it alleviates the need for labor-intensive dictionary creation, enhancing the efficiency and accuracy of ASR systems, and making them more applicable to real-world scenarios. For future enhancements, we recommend the inclusion of dialectal and spontaneous data in the dataset to broaden the model's adaptability. Additionally, fine-tuning the model for specific tasks can optimize its performance for targeted objectives or domains, further enhancing its effectiveness in those areas.

Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) have been utilised to forecast the foreign exchange market (Forex). However, such models usually exhibit unstable behaviour, as data perturbations often downgrade the functionality of the entire network due to their monolithic architecture. Hence, this study proposes a novel neuroscience-informed modular network applied to the closing prices and the sentiments retrieved from Yahoo Finance and Twitter APIs, aiming to anticipate better price fluctuations in Euro to British Pound Sterling (EUR/GBP) rather than monolithic methods. The proposed model is based on a new modular CNN, replacing pooling layers with orthogonal kernel initialisation RNNs coupled with Monte Carlo Dropout (MCD), namely MCoRNNMCD. It combines two modules: i) a convolutional simple RNN and ii) a convolutional Gated Recurrent Unit (GRU), where orthogonality and MCD are added to reduce the overfitting, assessing each module's uncertainty. These parallel feature extraction modules concatenate their outputs to a final three-layer Artificial Neural Network (ANN) decision-making module. A comprehensive comparison viewing objective evaluation metrics such as the Mean Square Error (MSE) proved that the proposed MCoRNNMCD-ANN outperformed single CNN, LSTM, GRU, and the state-of-the-art hybrid BiCuDNNLSTM, CLSTM, CNN-LSTM, and LSTM-GRU in forecasting hourly EUR/GBP closing price fluctuations.

Computer-aided image diagnostics (CAD) have been used in many fields of diagnostic medicine. It relies heavily on classical computer vision and artificial intelligence. Quantum neural network (QNN) has been introduced by many researchers around the world and presented recently by research corporations such as Microsoft, Google, and IBM. In this paper, the investigation of the validity of using the QNN algorithm for machine-based breast cancer detection was performed. To validate the learnability of the QNN, a series of learnability tests were performed alongside with classical convolutional neural network (CCNN). QNN is built using the Cirq library to perform the assimilation of quantum computation on classical computers. Series of investigations were performed to study the learnability characteristics of QNN and CCNN under the same computational conditions. The comparison was performed for real Mammogram data sets. The investigations showed success in terms of recognizing the data and training. Our work shows better performance of QNN in terms of successfully training and producing a valid model for smaller data set compared to CCNN.

Optical artificial neural networks (ONNs) — analog computing hardware tailored for machine learning [1,2] — have significant potential for ultra-high computing speed and energy efficiency [3]. We propose a new approach to architectures for ONNs based on integrated Kerr micro-comb sources [4] that is programmable, highly scalable and capable of reaching ultra-high speeds. We experimentally demonstrate the building block of the ONN — a single neuron perceptron — by mapping synapses onto 49 wavelengths of a micro-comb to achieve a high single-unit throughput of 11.9 Giga-FLOPS at 8 bits per FLOP, corresponding to 95.2 Gbps. We test the perceptron on simple standard benchmark datasets — handwritten-digit recognition and cancer-cell detection — achieving over 90% and 85% accuracy, respectively. This performance is a direct result of the record small wavelength spacing (49GHz) for a coherent integrated microcomb source, which results in an unprecedented number of wavelengths for neuromorphic optics. Finally, we propose an approach to scaling the perceptron to a deep learning network using the same single micro-comb device and standard off-the-shelf telecommunications technology, for high-throughput operation involving full matrix multiplication for applications such as real-time massive data processing for unmanned vehicle and aircraft tracking.

This article presents a detailed examination of the methodology and modeling tools utilized to analyze gas flow in pipelines, rooted in the fundamental principles of gas dynamics. The methodology integrates numerical simulations with modern neural network techniques, particularly focusing on the Physics-Informed Neural Network (PINN) method. This innovative approach combines artificial neural networks (ANNs) with physical equations, offering a more efficient and accurate way to model various complex processes and phenomena. The proposed mathematical model, based on the Euler equation, has been meticulously implemented using the Python language. Verification with analytical solutions ensures the accuracy and reliability of the computations. In the research, a comprehensive comparative analysis was conducted between results obtained using the PINN method and those from conventional Computational Fluid Dynamics (CFD) approaches. The analysis highlighted the advantages of the PINN method, which produced smoother pressure and velocity fluctuation profiles while reducing computation time, demonstrating its potential as a transformative modeling tool. The data derived from this study are of paramount importance for ensuring ongoing energy supply reliability and can also be used to create predictive models related to gas behavior in pipelines. The application of modeling techniques for gas flow simulation has the potential to revolutionize the integrity of our energy infrastructure and utilization of gas resources. However, it is crucial to emphasize that the effectiveness of such models relies on continuous monitoring and frequent updates to ensure alignment with real-world conditions. This research not only contributes to a deeper understanding of compressible gas flows but also underscores the crucial role of advanced modeling methodologies in the sustainable management of gas resources for both current and future generations.

The application of artificial intelligence to point-of-care testing (POCT) disease detection has become a hot research field, in which breath detection which detects the patient's exhaled VOCs combined with sensor arrays of Convolutional Neural Network (CNN) algorithms as a new lung cancer detection is attracting more researchers' attention. However, due to low accuracy, high complexity computation and large number of parameters make the CNN algorithms difficult transplant to the embedded system of POCT devices. A lightweight neural network (LTNet) in this work is proposed to deal with this problem, and meanwhile, achieve high-precision classification of acetone and ethanol gases, which are respiratory markers for lung cancer patients. Compared to currently popular lightweight CNN models such as EfficientNet, LTNet has fewer parameters (32K) and its training weight size is only 0.155 MB. LTNet achieved an overall classification accuracy of 99.06% and 99.14% in the own mixed gas dataset and the University of California (UCI) dataset, which are both higher than that of the five existing models’, and it also offers the shortest training (844.38s and 584.67s) and inference time (23s and 14s) in the same validation set. Compared to the existing CNN models, LTNet is more suitable for resource-limited POCT devices.

This study aimed at exploring what role artificial intelligence techniques could play in the futural numerical analysis. In this paper, a convolutional neural network techniques based on modified loss function is proposed as a surrogate of finite element method(FEM). Several surrogate-based physics-informed neural networks(PINNs) are developed to solve representative boundary value problems based on elliptic partial differential equations (PDEs). Results from the proposed surrogate-based approach are in good agreement with ones from conventional FEM. It is found that modification of the loss function could improve the prediction accuracy of the neural network. It is indicated that to some extent the artificial intelligence technique could replace conventional numerical analysis as a great surrogate model.

Forest fire identification is important for forest resource protection. Effective monitoring of forest fires requires the deployment of multiple monitors with different viewpoints, while most traditional recognition models can only recognize images from a single source. By ignoring the information from images with different viewpoints, these models produce high rates of missed and false alarms. In this paper, we propose a graph neural network model based on the similarity of dynamic features of multi-view images to improve the accuracy of forest fire recognition. The input features of the nodes on the graph are converted into relational features of different gallery pairs by establishing pairs (nodes) representing different viewpoint images and gallery images. The new feature library relationship is used to update the image gallery with dynamic features in order to achieve the estimation of similarity between images and improve the image recognition rate of the model. In addition, to reduce the complexity of image pre-processing process and extract key features in images effectively, this paper also proposes a dynamic feature extraction method for fire regions based on image segment ability. By setting the threshold value of HSV color space, the fire region is segmented from the image, and the dynamic features of successive frames of the fire region are extracted. The experimental results show that, compared with the baseline method Resnet, this paper's method is more effective in identifying forest fires, and its recognition accuracy is improved by 2%. And the scheme of this paper can adapt to different forest fire scenes, with better generalization ability and anti-interference ability.

Characterization of cancer cells and neural stem cells indicates that tumorigenicity and pluripotency are coupled cell properties determined by neural stemness, and tumorigenesis represents a process of progressive loss of original cell identity and gain of neural stemness. This reminds of a most fundamental process required for the development of the nervous system and body axis during embryogenesis, i.e., embryonic neural induction. Neural induction is that, in response to extracellular signals that are secreted by the Spemann-Mangold organizer in amphibians or the node in mammals and inhibit epidermal fate in ectoderm, the ectodermal cells lose their epidermal fate and assume the neural default fate and consequently, turn into neuroectodermal cells. They further differentiate into the nervous system and also some non-neural cells via interaction with adjacent tissues. Failure in neural induction leads to failure of embryogenesis, and ectopic neural induction due to ectopic organizer or node activity or activation of embryonic neural genes causes a formation of secondary body axis or a conjoined twin. During tumorigenesis, cells progressively lose their original cell identity and gain of neural stemness, and consequently, gain of tumorigenicity and pluripotency, due to various intra-/extracellular insults in cells of a postnatal animal. Tumorigenic cells can be induced to differentiation into normal cells and integrate into normal embryonic development within an embryo. However, they form tumors and cannot integrate into animal tissues/organs in a postnatal animal because of lack of embryonic inducing signals. Combination of studies of developmental and cancer biology indicates that neural induction drives embryogenesis in gastrulating embryos but a similar process drives tumorigenesis in a postnatal animal. Tumorigenicity is the manifestation of aberrant occurrence of pluripotent state in a postnatal animal. Pluripotency and tumorigenicity are both but different manifestations of neural stemness in pre- and postnatal stage, respectively, of animal life. The unique property of neural stemness is derived from the evolutionary advantage of neural genes and the neural-biased state of the last common unicellular ancestors of metazoan. Based on these findings, I discuss about some confusion in cancer research, propose to distinguish the causality and associations and discriminate causal and supporting factors involved in tumorigenesis, and suggest revisiting the focus of cancer research.

This paper introduces a novel architecture—bidirectional optical neural network (BONN)—for providing backward connections alongside forward connections in artificial neural networks (ANNs). BONN incorporates laser diodes and photodiodes and exploits the properties of Köhler illumination to establish optical channels for backward directions. Thus, it has bidirectional functionality that is crucial for algorithms such as the backpropagation algorithm. BONN has a scaling limit of 96 × 96 for input and output arrays, and a throughput of 8.5 × 1015 MAC/s. While BONN's throughput may rise with additional layers for continuous input, limitations emerge in the backpropagation algorithm as its throughput doesn't scale with layer count. Successful BONN-based implementation of the backpropagation algorithm requires the development of fast spatial light modulator to accommodate frequent data flow changes. A two-mirror-like BONN and its cascaded extension are alternatives for multilayer emulation, and they help save hardware space and increase the parallel throughput for inference. An investigation into the application of the clustering technique to BONN revealed its potential to help overcome scaling limits and to provide full interconnections for backward directions between doubled input and output ports. BONN's bidirectional nature holds promise for enhancing supervised learning in ANNs and increasing hardware compactness.

Artificial Intelligence (AI) made its first appearance in 1956, and since then it has progressively introduced itself in healthcare system and patients’ information and care. AI functions can be grouped under the following headings: Machine Learning (ML), Deep Learning (DL), Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Computer Vision (CV). Chronic limb-threatening ischemia (CLTI) represents the last stage of peripheral artery disease (PAD), and has increased over recent years, together with the rise in prevalence of diabetes and population ageing. Nowadays AI grants the possibility of developing new diagnostic and treatment solutions in the vascular field, given the possibility of accessing clinical, biological, and imaging data. By assessing the vascular anatomy in every patient, as well as the burden of atherosclerosis, and classifying the level and degree of disease, sizing and planning the best endovascular treatment, defining the perioperative complications risk, integrating experiences and resources between different specialties, identifying latent PAD, thus offering evidence-based solutions and guiding surgeons in the choice of the best surgical technique, AI challenges the role of the physician’s experience in PAD treatment.

Partial differential equations (PDEs) are used to describe a wide range of phenomena, such as heat transfer, fluid dynamics, and quantum mechanics. By solving PDEs, we can ob- tain insights into the behavior of the system and make predictions about its future evolution. Conventional numerical methods for obtaining the approximate solutions of PDEs may re- quire extensive computational resources and time, especially for complex PDEs and large- scale problems. The recently developed physics-informed neural network (PINN) has shown promise in a variety of scientific and engineering fields by incorporating physical laws into the loss functions of the neural network (NN). In addition, the training of PINN does not require ground truth data, but it has poor generalization ability to unseen domains. On the other hand, a convolutional neural network (CNN) has fast inference and better generalization ability, but it requires a large amount of training data. Taking the advantages of PINN and CNN by using Legendre multiwavelets (LMWs) as basis functions, we introduce a new method to approach the PDEs in this paper, namely Physics-Informed Legendre Multiwavelets CNN (PiLMWs-CNN), in order to continuously approximate a grid-based state representation that can be handled by a CNN. PiLMWs-CNN enable us to train our models using only physics-informed loss functions without any pre- computed training data, simultaneously providing fast and continuous solutions that gener- alize to previously unknown domains. In particular, the LMWs can simultaneously possess compact support, orthogonality, symmetry, high smoothness, and high approximation order. Compared to orthonormal polynomial (OP) bases, the approximation accuracy can be greatly increased and computation costs can be significantly reduced by using LMWs. We applied PiLMWs-CNN to approximate the damped wave equation, incompressible Navier-Stokes (N- S) equation, and two-dimensional heat conduction equation. The experimental results show that this method provides more accurate, efficient, and fast convergence with better stability when approximating the solution of PDEs.

Human activity recognition (HAR) using wearable sensors is gaining popularity in the healthcare domain, enabling continuous monitoring and timely interventions. The conventional HAR systems are centralised and need data sharing with a server for processing and model training. However, this data sharing leads to various challenges, such as privacy concerns, increased communication and storage costs, inefficient computational algorithms, data scarcity due to unwillingness to share, and difficulty achieving high accuracy and personalised models simultaneously. To overcome these challenges, this paper introduces a privacy and energy-aware federated learning (FL) framework for model training without data sharing. The proposed scheme incorporates spiking neural networks (SNNs), which are particularly suited for resource-constrained devices due to their event-driven information processing. The key idea is to synergise the strengths of SNNs and long short-term memory (LSTM) to create a hybrid model for robust HAR. The proposed hybrid spiking LSTM (S-LSTM) model uses the LSTM as an input layer to capture temporal dependencies efficiently, followed by spiking layers that replace the typical activation using leaky integrate-and-fire neurons. We employ the combination of surrogate gradient learning and backpropagation through time (BPTT) to update the weights of SNN layers in a supervised manner. The performance of the proposed model is evaluated using two publicly available datasets for both indoor and outdoor scenarios. The results show that the proposed S-LSTM performs much better than spike convolutional neural networks (S-CNN) and simple CNN, achieving the accuracy of 97% and 87% for indoor and outdoor datasets, respectively. Personalisation, also known as fine-tuning the global model using local data, is crucial in HAR. It allows the model to adapt to individual behaviour and preferences, making predictions more accurate. The results also show that personalisation improves the performance of participants up to 9% on average. Additionally, analysis with 50% random client participation reduces communication costs by 50% with minimal impact on accuracy.

A scalable optical convolutional neural network (SOCNN) based on free-space optics and Koehler illumination was proposed to address the limitations of the previous 4f correlator system. Unlike Abbe illumination, Koehler illumination provides more uniform illumination and reduces crosstalk. SOCNN allows for scaling up of the input array and the use of incoherent light sources. Hence, the problems associated with 4f correlator systems can be avoided. We analyzed the limitations in scaling the kernel size and parallel throughput and found that SOCNN can offer a multilayer convolutional neural network with massive optical parallelism.

This study uses machine learning to predict the convergence results of the Duffing equation with and without damping. The Duffing equation represents a nonlinear second-order differential equation with interesting behavior in undamped free vibration and forced vibration with damping. Convergence alternates randomly between 1 and -1 in undamped free vibration, depending on initial conditions. For forced vibration with damping, multiple factors influence vibration patterns. We utilize the fourth-order Runge-Kutta method to collect convergence results for both conditions. Machine learning techniques, specifically the long short term memory (LSTM) and LSTM-Neural Network (LSTM-NN) method, are employed to predict these convergence values. The LSTM-NN model is a hybrid approach that combines the LSTM method with the addition of hidden layers of neurons. Both the LSTM and LSTM-NN models are thoroughly explored and analyzed in this research. The research process involves three stages: data preprocessing, training, and verification. The results show that the LSTM-NN model becomes more adept at predicting binary datasets, boasting an impressive accuracy of up to 98%. However, when it comes to predicting multiple solutions, the traditional LSTM method outperforms the LSTM-NN approach.

Convolutional neural networks (CNNs), inspired by biological visual cortex systems, are a powerful category of artificial neural networks that can extract the hierarchical features of raw data to greatly reduce the network parametric complexity and enhance the predicting accuracy. They are of significant interest for machine learning tasks such as computer vision, speech recognition, playing board games and medical diagnosis [1-7]. Optical neural networks offer the promise of dramatically accelerating computing speed to overcome the inherent bandwidth bottleneck of electronics. Here, we demonstrate a universal optical vector convolutional accelerator operating beyond 10 Tera-FLOPS (floating point operations per second), generating convolutions of images of 250,000 pixels with 8-bit resolution for 10 kernels simultaneously — enough for facial image recognition. We then use the same hardware to sequentially form a deep optical CNN with ten output neurons, achieving successful recognition of full 10 digits with 900 pixel handwritten digit images with 88% accuracy. Our results are based on simultaneously interleaving temporal, wavelength and spatial dimensions enabled by an integrated microcomb source. This approach is scalable and trainable to much more complex networks for demanding applications such as unmanned vehicle and real-time video recognition.

Speech recognition, also known as automatic speech recognition (ASR), is a technology that enables software to transcribe spoken language into text. However, existing Amharic ASR methods require multiple separate blocks, such as language, acoustic, and pronunciation models with dictionaries, which can be time-consuming and influence performance. This study proposes an approach that replaces much of the speech pipeline with a single recurrent neural network (RNN) architecture. Our proposed architecture is based on a hybrid approach that combines a convolutional neural network (CNN) with a recurrent neural network (RNN) and a connectionist temporal classification (CTC) loss function. We conducted several experiments with noisy audio data that contain 20,000 valid sentences. The model was evaluated using the word error rate (WER) metric, achieving impressive results of 7% WER on noisy data. This approach has significant implications for the field of speech recognition, as it reduces the human effort required to create dictionaries and improves the efficiency and accuracy of ASR systems, making them more practical for real-world applications.

Creating self-learning algorithms, developing deep neural networks and improving other methods that "learn" for various areas of human activity is the main goal of the theory of machine learning. It helps to replace the human with a machine, aiming to increase the quality of production. The theory of artificial neural networks, which already have replaced the humans in problems of detection of moving objects, recognition of images or sounds, time series prediction, big data analysis and numerical methods remains the most dispersed branch of the theory of machine learning. Certainly, for each area of human activity it is necessary to select appropriate neural network architectures, methods of data processing and some novel tools from applied mathematics. But the universal problem for all these neural networks with specific data is the achieving the highest accuracy in short time. Such problem can be resolved by increasing sizes of architectures and improving data preprocessing, where the accuracy rises with the training time. But there is a possibility to increase the accuracy without time growing, applying optimization methods. In this survey we demonstrate existing optimization algorithms of all types, which can be used in neural networks. There are presented modifications of basic optimization algorithms, such as stochastic gradient descent, adaptive moment estimation, Newton and quasi-Newton optimization methods. But the most recent optimization algorithms are related to information geometry, for Fisher-Rao and Bregman metrics. This approach in optimization extended the theory of classic neural networks to quantum and complex-valued neural networks, due to geometric and probabilistic tools. There are provided applications of all introduced optimization algorithms, what delighted many kinds of neural networks, which can be improved by including any advanced approaches in minimization of the loss function. Afterwards, we demonstrated ways of developing optimization algorithms in further researches, engaging neural networks with progressive architectures. Classical gradient based optimizers can be replaced by fractional order, bilevel and, even, gradient free optimization methods. There is a possibility to add such analogues in graph, spiking, complex-valued, quantum and wavelet neural networks. Besides the usual problems of image recognition, time series prediction, object detection, there are many are other tasks for modern theory of machine learning, such as solving problem of quantum computations, partial differential and integro-differential equations, stochastic processes and Brownian motion, making decisions and computer algebra.

In the pursuit of energy efficiency and reduced environmental impact, adequate ventilation in enclosed spaces is essential. This study presents a hybrid neural network model designed for real-time monitoring and prediction of environmental variables. The system comprises two phases: An IoT hardware-software platform for data acquisition and decision-making, and a hybrid model combining short-term memory and convolutional recurrent structures. The results are promising and hold potential for integration into parallel processing AI architectures.

In computer vision and image analysis, Convolutional Neural Networks (CNNs) and other deep learning models are at the forefront of research and development. These advanced models have proven to be highly effective in tasks related to computer vision. One technique that has gained prominence in recent years is the construction of ensembles using Deep CNNs. These ensembles typically involve combining multiple pre-trained CNNs to create a more powerful and robust network. The purpose of this study is to evaluate the effectiveness of building CNN ensembles by combining several advanced techniques. Tested here are CNN ensembles constructed by replacing ReLU layers with different activation functions, employing various data augmentation techniques, and utilizing several algorithms, including some novel ones, that perturb network weights. Experimental results performed across many data sets representing different tasks demonstrate that our proposed methods for building deep ensembles produces superior results. All the resources required to replicate our experiments are available at https://github.com/LorisNanni.

We present a simplified computational rule for the back-propagation formulas for artificial neural networks. In this work, we provide a generic two-step rule for the back-propagation algorithm in matrix notation. Moreover, this rule incorporates both the forward and backward phases of the computations involved in the learning process. Specifically, this recursively computing rule permits the propagation of the changes to all synaptic weights in the network, layer by layer, efficiently. In particular, we use this rule to compute both the up and down partial derivatives of the cost function of all the connections feeding into the output layer.

Numerous approaches address over-fitting in neural networks: by imposing a penalty on the parameters of the network (L1, L2, etc); by changing the network stochastically (drop-out, Gaussian noise, etc.); or by transforming the input data (batch normalization, etc.). In contrast, we aim to ensure that a minimum amount of supporting evidence is present when fitting the model parameters to the training data. This, at the single neuron level, is equivalent to ensuring that both sides of the separating hyperplane (for a standard artificial neuron) have a minimum number of data points — noting that these points need not belong to the same class for the inner layers. We firstly benchmark the results of this approach on the standard Fashion-MINST dataset, comparing it to various regularization techniques. Interestingly, we note that by nudging each neuron to divide, at least in part, its input data, the resulting networks make use of each neuron, avoiding a hyperplane completely on one side of its input data (which is equivalent to a constant into the next layers). To illustrate this point, we study the prevalence of saturated nodes throughout training, showing that neurons are activated more frequently and earlier in training when using this regularization approach. A direct consequence of the improved neuron activation is that deep networks are now easier to train. This is crucially important when the network topology is not known a priori and fitting often remains stuck in a suboptimal local minima. We demonstrate this property by training a network of increasing depth (and constant width): most regularization approaches will result in increasingly frequent training failures (over different random seeds) whilst the proposed evidence-based regularization significantly outperforms in its ability to train deep networks.

Background: Disruption of alternative splicing (AS) is frequently observed in cancer and it might represent an important signature for tumor progression and therapy. Exon skipping (ES) represents one of the most frequent AS events and in non-small cell lung cancer (NSCLC) MET exon 14 skipping was shown to be targetable. Methods: We constructed a neural network (NN) specifically designed to detect MET exon 14 skipping events using RNAseq data. Furthermore, for discovery purpose we also developed a sparsely connected autoencoder to identify uncharacterized MET isoforms. Results: The NN had 100% Met exon 14 skipping detection rate, when tested on a manually curated set of 690 TCGA bronchus and lung samples. When globally applied to 2605 TCGA samples, we observed that the majority of false positives was characterized by a blurry coverage of exon 14, but interesting they share a common coverage peak in the second intron and we speculate that this event could be the transcription signature of a LINE1-MET fusion. Conclusions: Taken together our results indicate that neural networks can be an effective tool to provide a quick classification of pathological transcription events and sparsely connected autoencoders could represent the basis for the development of an effective discovery tool.

Optical artificial neural networks (ONNs) have significant potential for ultra-high computing speed and energy efficiency. We report a novel approach to ONNs that uses integrated Kerr optical micro-combs. This approach is programmable and scalable and is capable of reaching ultra-high speeds. We demonstrate the basic building block ONNs — a single neuron perceptron — by mapping synapses onto 49 wavelengths to achieve an operating speed of 11.9 x 10⁹ operations per second, or Giga-OPS, at 8 bits per operation, which equates to 95.2 gigabits/s (Gbps). We test the perceptron on handwritten-digit recognition and cancer-cell detection — achieving over 90% and 85% accuracy, respectively. By scaling the perceptron to a deep learning network using off-the-shelf telecom technology we can achieve high throughput operation for matrix multiplication for real-time massive data processing.

Real-world time series data sets contain a combination of linear and nonlinear patterns, making the time series forecasting problem more challenging. In this paper, a new hybrid methodology is introduced for forecasting univariate time series data sets using a multiplicative error modeling approach. An autoregressive integrated moving average (ARIMA) model is combined with an autoregressive neural network (ARNN) for improving the predictions of individual forecast models. The proposed multiplicative ARIMA-ARNN model glorifies the chances of capturing the different combinations of linear and nonlinear patterns in time series. The model shows outstanding performance on six standard time-series data sets compared to other widely used single and hybrid forecasting models.

Using smartphone especially android platform has already got eighty percent market shares, due to aforementioned report, it becomes attacker’s primary goal. There is a growing number of private data onto smart phones and low safety defense measure, attackers can use multiple way to launch and to attack user’s smartphones.(e.g. Using different coding style to confuse the software of detecting malware). Existing android malware detection methods use multiple features, like safety sensor API, system call, control flow structure and data information flow, then using machine learning to check whether its malware or not. These feature provide app’s unique property and limitation, that is to say, from some perspectives it might suit for some specific attack, but wouldn’t suit for others. Nowadays most malware detection methods use only one aforementioned feature, and these methods mostly analysis to detect code, but facing the influence of malware’s code confusion and zero-day attack, aforementioned feature extraction method may cause wrong judge. So, it’s necessary to design an effective technique analysis to prevent malware. In this paper, we use the importance of word from apk, because of code confusion, some malware attackers only rename variables, if using general static analysis wouldn’t judge correctly, then use these importance value to go through our proposed method to generate picture, finally using convolutional neural network to see whether the apk file is malware or not.

Deep learning is emerging in many industrial sectors in hand with big data analytics to streamline production. In the biomanufacturing sector, big data infrastructure is lagging comparatively to other industries. A promising approach is to combine Deep Neural Networks (DNN) with prior knowledge in Hybrid Neural Network (HNN) workflows that are less dependent on the quality and quantity of data. This paper reviews published articles over the past 30 years on the topic of HNN applications to bioprocesses. It revealed that HNNs were applied to various bioprocesses, including microbial cultures, animal cells cultures, mixed microbial cultures, and enzyme biocatalysis. HNNs were mainly applied for process analysis, process monitoring, development of software sensors, open- and closed-loop control, batch-to-batch control, model predictive control, intensified design of experiments, quality-by-design, and recently for the development of digital twins. Most previous HNN studies combined shallow Feedforward Neural Networks (FFNNs) with physical laws, such as macroscopic material balance equations, following the semiparametric design principle. Only recently, deep HNNs based on deep FFNNs, Convolution Neural Networks (CNN), Long Short-Term Memory (LSTM) networks and Physics Informed Neural Networks (PINNs) have been reported. The biopharma sector is currently a major driver but applications to biologics quality attributes, new modalities, and downstream processing are significant research gaps.

Intracortical microelectrode arrays (MEAs) can be used in a range of applications, from basic neuroscience research to providing an intimate interface with the brain as part of a Brain-Computer Interface (BCI) system aimed to restore function for people living with neurological disorders or injuries. Unfortunately, MEAs tend to fail prematurely, leading to a loss in functionality for many applications. An important contributing factor in MEA failure is oxidative stress resulting from chronically inflammatory-activated microglia and macrophages releasing reactive oxygen species (ROS) around the implant site. Antioxidants offer a means for mitigating oxidative stress and improving tissue health and MEA performance. Here, we investigate using the clinically available antioxidant dimethyl fumarate (DMF) to reduce the neuroinflammatory response and improve MEA performance in a rat MEA model. Daily treatment of DMF for 16 weeks resulted in a significant improvement in the recording capabilities of MEA devices during the sub-chronic (Weeks 5-11) phase (42% active electrode yield vs. 35% for control). However, these sub-chronic improvements were lost in the chronic implantation phase, as a more exacerbated neuroinflammatory response occurs in DMF-treated animals by 16 weeks post-implantation. Yet, neuroinflammation was indiscriminate between treatment and control groups during the sub-chronic phase. Although worse for chronic use, a temporary improvement (<12 weeks) in MEA performance is meaningful. Providing short-term improvement to MEA devices using DMF can allow for improved use for limited-duration studies. Further efforts should be taken to explore the mechanism behind a worsened neuroinflammatory response at the 16-week time point for DMF-treated animals to assess its usefulness to specific applications.

Abstract: It has been shown that three-dimensional self-assembled multicellular structures derived from human pluripotent stem cells show electrical activity similar to EEG. More recently, neurons were successfully embedded in a digital game-worlds (Kagan, et al. 2022). The biologically inspired neural network (BNN), expressing human cortical cells, was able to show internal modification and learn the task at hand (predicting the trajectory of a digital ball while moving a digital paddle). In other words, the system allowed to read motor information and write sensory data into cell cultures. In this article, we discuss Neural Correlates of Consciousness (NCC) theories, and their capacity to predict or even allow for consciousness in a BNN. We found that In-formation Integration Theory (IIT) is the only NCC that offers the possibility for a BNN to show consciousness. IIT argues that any system capable of integrating information will have some degree of phenomenal consciousness. The pattern of activity appearing in the BNN, with increased density of sensory information leading to better performance, implies that the BNN could be conscious. This may have profound implications from a physiological, philosophical, and ethical perspective.

This article presents a forecast model that uses a hybrid architecture of recurrent neural networks (RNN) with surface neural networks (ANN), based on historical records of exported active energy (EAE) and weather data. Two types of models were developed: the first type includes six models that use EAE records and weather variables as inputs, while the second type includes eight models that use only weather variables. Different metrics were applied to assess the performance of these models, and the best model of each type was selected. Finally, a comparison of the performance between the selected models of both types is presented, and they are validated with real data provided by a solar plant, achieving acceptable levels of accuracy. The selected model of the first type has an RMSE of 0.19, MSE of 0.03, MAE of 0.09, a correlation coefficient of 0.96, and a determination coefficient of 0.93. The other selected model of the second type showed lower precision in the metrics (RMSE = 0.24, MSE = 0.06, MAE = 0.10, Corr. Coef. = 0.95, and Det. Coef. = 0.90). Both models demonstrated good performance and acceptable accuracy in forecasting the weekly photovoltaic energy production of the solar plant.

Sense element engagement theory explains how neural networks produce cortical prosthetic vision. A major prediction of the theory can be tested by developing a device which is expected to enable perception of continuous forms in altered visual geometries. The research reported here completes several essential steps in developing this device: (1) replication of simulations that are consistent with the theory using the NEST simulator, which can also be used for full-scale network emulation by a neuromorphic computer; (2) testing whether results consistent with the theory survive increasing the scale and duration of simulations; (3) establishing a method that uses numbers of spikes produced by network neurons to report the number of phosphenes produced by cortical stimulation; and (4) simulating essential functions of the prosthetic device. NEST simulations replicated early results and increasing their scale and duration produced results consistent with the theory. A decision function created using multinomial logistic regression correctly classified the expected number of phosphenes for 2080 spike number distributions for each of three sets of data, half of which arise from simulations expected to yield continuous visual forms on an altered visual geometry. A process for modulating electrical stimulation amplitude based on intermittent population recordings that is predicted to produce continuous visual forms was successfully simulated. The classification function developed using logistic regression will be used to tune this process as the scale of simulations is further increased.