Abstract: Text-attributed graph node classification is still a challenge since it needs to reason about the topology structure simultaneously with the free-text semantics. Although graph neural network can perform well on structural propagation,they tend to be blind for the details in the text associated with nodes. On the other hand, LLMs have excellent NLU skills and are weak on structured,multi-hop reasoning over network agents.To address the above gap, in this work we propose FinSCRA, a novel LLM-powered multi-chain reasoning framework to inject domain-aware reasoning capability into a financial LLM with parameters efficient fine-tuning. Specifically, our framework designs a hierarchy of structured reasoning chains (single-hint,parallel, cascaded, and hybrid methods to extract and fuse the semantic signals like sentiment, correlation, and risk signals in the nodes’ text.A fusion layer based on fuzzy logic fuses the results of different reasoning lines for better robustness and explainability.While FinSCRA is generic and can be applied to other types of text-attributed graphs, here we assess its performance on credit risk analysis in supply chain networks,on the task of entity relation extraction, in which entities are related through their financial relation and described with rich text reports; we show experimentally on realworld datasets that our model FinSCRA greatly outperforms graphbased as well as LLM-based baselines,as an accurate and explainable technique to perform node classification over complex networked systems.We release our code and models for further research on LLM-grap.

Abstract: In recommendation systems, representing user-item interactions as a bipartite network is a fundamental approach that provides a structured way to model relationships between users and items, allowing for efficient predictions via network science. Collaborative filtering is one of the most widely used and actively researched techniques for recommendation systems, its rationale is to predict user preferences based on shared patterns in user interactions, and vice versa. Memory-based collaborative filtering relies on directly analyzing user-item interactions to provide recommendations using similarity measures, and differs from model-based collaborative filtering which builds a predictive model using machine learning techniques such as neural networks. With the rise of machine learning, memory-based collaborative filtering has often been overshadowed by model-based approaches. However, the recent success of SSCF, a newly proposed memory-based method, has renewed interest in the potential of memory-based approaches. In this paper, we propose Network Shape Automata (NSA), a memory-based collaborative filtering method grounded in the connectivity shape of the bipartite network topology. NSA leverages the Cannistraci-Hebb theory proposed in network science to define brain-inspired network automata, using this paradigm as the foundation for its similarity measure. We evaluate NSA against a range of advanced collaborative filtering methods, both memory-based and model-based, across 16 bipartite network datasets spanning complex systems domains such as social networks and biological networks. Results show that NSA consistently achieves strong performance across diverse datasets and evaluation metrics, ranking most often first on average. Notably, NSA demonstrates strong robustness to network sparsity, while preserving the simplicity, interpretability, and training-free nature of memory-based methods. As a pioneering effort to bridge link prediction and recommendation tasks, NSA not only highlights the untapped potential of memory-based collaborative filtering but also demonstrates the effectiveness of the Cannistraci-Hebb theory in modeling network evolution within recommendation systems.

Abstract: Background/Objectives: The rapid advancement of artificial intelligence (AI) has had a notable impact in the healthcare field, particularly in the realm of assessment and diagnosis. One specific area where the integration of AI technologies shows promise is the evaluation of progressive neurological disorders (PNDs). PNDs are characterized by a progressive decline in neurological function, resulting in changes in cognition, movement, and communication. PNDs pose significant challenges in terms of early detection and categorization. Speech and voice changes are important clinical markers in many PNDs. Therefore, the utilization of AI applications for the analysis and classification of speech and voice samples could prove beneficial for streamlining the diagnostic process. This systematic review aimed to investigate the current utilization of AI in the assessment and diagnosis of PNDs through speech signal analysis over the past decade. Methods: In adherence to PRISMA guidelines, Scopus, PubMed, and Web of Science were searched for studies related to machine learning (ML) and deep learning (DL) for speech and voice assessment in people with PNDs. Results: A total of 102 studies were identified for inclusion between 2013 and 2023. The reviewed studies demonstrated a wide range of accuracy, with reported values ranging from 67.43% to 99%. Support Vector Machines (SVMs) were the most frequently used ML models across studies, demonstrating reliable performance in both speech and voice data analysis. Conclusions: AI-based analysis of speech and voice shows strong potential as a non-invasive tool for supporting the assessment and diagnosis of PNDs. The high accuracy reported across studies highlights the promise of these approaches, although methodological variability underscores the need for greater standardization and clinical validation.

Abstract: In this paper, we move beyond the classical setting by redefining the Chebyshev functional in the context of q-circles situated within Minkowski space, rather than the standard Euclidean circles in R². This approach introduces a new theoretical framework suitable for non-Euclidean geometries. We derive sharp estimates for the functional when applied to functions on q-circles that adhere to Hölder-type continuity conditions.

Abstract: In many cases, the pieces of information at our disposal come from a recommender source, that can be either an official news system, a large language model or simply a social network. Often, also, these messages are build so to promote their active spreading, which, on the other hand, has a positive effect on one’s own popularity. However, the content of the message can be false, giving origin to a phenomenon analogous to the spreading of a disease. In principle, there is always the possibility of checking the correctness of the message by “investing” some time, so we can say that this checking has a cost. We develop a simple model based on the mechanism of “risk perception” (propensity of checking the falseness of a message) and mutual trustability, based on the average number of fake messages received and checked. On the other side, the probability of emitting a fake message is inversely proportional to risk perception and the affinity (trustability) among agents is also exploited by the recommender system. This model represents an integration of cognitive psychology with computational agent-based modeling.

Abstract: Artificial neural networks (ANNs) achieve remarkable performance but at the unsustainable cost of extreme parameter density. In contrast, biological networks operate with ultra-sparse, highly organized structures, where dendrites play a central role in shaping information integration. Here we introduce the Dendritic Network Model (DNM), a generative framework that bridges this gap by embedding dendritic-inspired connectivity principles into sparse artificial networks. Unlike conventional random initialization, DNM defines connectivity through parametric distributions of dendrites, receptive fields, and synapses, enabling precise control of modularity, hierarchy, and degree heterogeneity. This parametric flexibility allows DNM to generate a wide spectrum of network topologies, from clustered modular architectures to scale-free hierarchies, whose geometry can be characterized and optimized with network-science metrics. Across image classification benchmarks (MNIST, Fashion-MNIST, EMNIST, CIFAR-10), DNM consistently outperforms classical sparse initializations at extreme sparsity (99\%), in both static and dynamic sparse training regimes. Moreover, when integrated into state-of-the-art dynamic sparse training frameworks and applied to Transformer architectures for machine translation, DNM enhances accuracy while preserving efficiency. By aligning neural network initialization with dendritic design principles, DNM demonstrates that sparse bio-inspired network science modelling is a structural advantage in deep learning, offering a principled initialization framework to train scalable and energy-efficient machine intelligence.

Abstract: Multi-agent systems often rely on long-term memory or shared knowledge bases to enhance collaborativeefficiency, yet this introduces risks of memory poisoning and cross-agent propagation. Addressing the covertdiffusion of poisoned information during collaboration, this study proposes a memory poisoning detection andrepair method tailored for multi-agent environments.This approach constructs an evidence graph based onmemory source credibility and content consistency to validate newly added memories. It combines contrastivelearning models to identify anomalous memories exhibiting command-induced characteristics. Upon detectingpoisoning, further propagation is suppressed through isolation, rewriting, and conflict resolution. Experimentsevaluated the method using 60 collaborative tasks, approximately 210,000 memory records, and 12,000injected poisoned samples.Results demonstrate an AUC of 0.94 in poisoning detection, reducing misbehaviorrates from 15.6% to 2.3% while decreasing cross-agent propagation by 78.1% on average, with minimal impacton overall task efficiency.

Abstract: We present a unified algebraic framework, the "Golay-Hopf Machine," which synthesizes four distinct mathematical structures: Golay coding theory, Hida theory, Iwasawa theory, and Yang-Baxter integrability. By defining a Hopf algebra structure on the binary Golay weights W = {0, 8, 12, 16, 24}, we show that: (1) Hida transitions correspond to the coproduct ∆, (2) Galois height corresponds to the counit ε, and (3) the weight complement w 7 → 24 − w acts as the antipode S satisfying S2 = id. We formally verify in Lean 4 that this structure satisfies the Yang-Baxter compatibility condition for heights and the Iwasawa logarithmic identity. All core algebraic results are verified with zero axioms and zero sorry statements. Finally, we sketch a roadmap for extending this framework to Anabelian geometry.

Abstract: The algebraic analysis of linear code parameters reveals deep connections with cryptographic constructions, including the information dispersal algorithms (IDAs) and secret-sharing schemes. In this work, we propose an algebraic method for constructing bases of binary linear codes from subsets of codewords selected according to their generalized Hamming weights (GHWs). The approach employs a degree-compatible monomial ordering on the polynomial ring F₂[x₁, . . . , x_n] and imposes the conditions d₁(C) = 1 and d_k (C) = n. Under these assumptions, we prove the existence of a generator matrix containing an invertible k × k submatrix, which guarantees correct information reconstruction. This structural property enables the direct application of binary linear codes to information dispersal and recovery mechanisms without the need for larger finite fields. We validate the proposed framework through algebraic proofs and an explicit example illustrating both the dispersal and recovery procedures. These results provide a theoretical foundation for the design of information dispersal schemes relying exclusively on binary linear codes.

Abstract: The literature indicates that the qubit requirements for factoring RSA-2048 remain on the order of 1 million, under commonly assumed architectures and error-correction models, leaving a substantial gap between current resource estimates and near-term practical feasibility. Reducing this requirement to the low thousands qubit regime therefore remains an important open research objective. This work proposes a hybrid classical-quantum algorithm using a classical modular exponentiation subroutine with a Quantum Number Theoretic Transform (QNTT) circuit, to increase the speed and reduce the number of quantum components, including gates and qubits, to factor integer numbers, which serve as keys in cryptographic methods, like RSA and ECC, when compared with Shor’s algorithm. Several composite numbers, the result of multiplication of two primes, were validated through both simulation and real quantum hardware by benchmarking the full Shor pipeline on simulation and on a real IBM quantum computer. In simulation, the proposed Jesse–Victor–Gharabaghi (JVG) algorithm achieved substantial practical reductions in computational resources, decreasing runtime from 174.1 s to 5.4 s, memory usage from 12.5 GB to 0.27 GB, and quantum gate counts by approximately 99%. Because Shor and JVG use different register sizes for the same composite N, the reported gate/depth reductions should be interpreted as end-to-end quantum-resource budgets to factor the same N, rather than a per-qubit or transform-only efficiency claim. On quantum hardware, JVG reduced the required runtime from 67.8 s to 2 s, and the quantum gate counts by over 98%. Projection for RSA-2048 indicates that the JVG algorithm significantly outperforms Shor’s approach, requiring a projected quantum runtime of 11 hours for a factorization under identical scaling assumptions. The results from these evaluations support JVG as a more hardware-compatible and robust noise-tolerant substitute for Shor’s framework, offering a viable path toward practical quantum integer factorization on near-term Noisy Intermediate-Scale Quantum (NISQ) devices.

Abstract: Nature-inspired metaheuristic algorithms are commonly applied to complex combinatorial optimization problems where exact methods are computationally impractical. Tourist route optimization is a representative multi-objective problem characterized by realistic constraints such as travel time, cost, opening hours, and transportation modes. Although Mushroom Reproduction Optimization (MRO) is computationally efficient, it often suffers from premature convergence in complex search spaces. This paper proposes a novel hybrid algorithm, Mushroom–Weed Hybrid Reproduction Optimization (MWHRO), which integrates the colony-based local search of MRO with the fitness-proportional reproduction and competitive elimination mechanisms of Invasive Weed Optimization (IWO). Hybridization enhances population diversity and global exploration while preserving fast convergence. The proposed algorithm is evaluated on a realistic tourist route optimization problem using real-world data from Zagreb, Croatia, across multiple transportation modes and objective-weight scenarios. Performance is compared against Ant Colony Optimization (ACO), IWO, and standard MRO under equal evaluation budgets. Experimental results demonstrate that MWHRO consistently achieves high-quality solutions with significantly lower execution times, particularly in constrained and multimodal scenarios. Statistical analysis confirms the robustness and practical suitability of the proposed approach for real-world route optimization.

Abstract: While ANN-to-SNN conversion is a pivotal approach to obtain SNNs, current methods mostly focus on dense architectures, disregarding the structural sparsity fundamental to brain neural networks. To bridge this gap, we propose a novel framework that integrates Cannistraci-Hebb Training (CHT)—a brain-inspired Dynamic Sparse Training algorithm—to instill biologically plausible topologies into SNNs. Through our framework, the converted SNNs directly inherit emergent brain-like properties, such as meta-depth and small-worldness, from the sparse ANNs. We confirm the brain-like topology trained by CHT and then investigate our framework across different conversion approaches. Our approach achieves comparable or superior accuracy to dense counterparts on both convolutional neural networks (CNNs) and Vision Transformer (ViT), while reducing theoretical energy consumption by over 60%. Empirically, we validate the framework's superiority over pruning baselines and direct SNN sparse training in terms of the accuracy-energy trade-off.

Abstract: We show that in a vector space over Z₃, the union of any four linear bases is an additive basis, thus proving the Additive Basis Conjecture for p=3 and providing an alternative proof of the weak 3-flow conjecture.

Abstract: Zero-shot generalization to out-of-distribution (OOD) teammates and opponents in open-ended multi-agent systems (MAS) remains a fundamental challenge for general-purpose AI. Existing multi-agent reinforcement learning (MARL) paradigms, such as self-play and population-based training, often collapse to a limited subset of Nash equilibria, leaving agents brittle when faced with semantically diverse, unseen behaviors. Recent approaches that invoke large language models (LLMs) at run time can improve adaptability but introduce substantial latency and can become less reliable as task horizons grow; in contrast, LLM-assisted reward-shaping methods remain constrained by the inefficiency of the inner reinforcement-learning loop. To address these limitations, we propose LLM-TOC (LLM-Driven Theory-of-Mind Adversarial Curriculum), which casts generalization as a bi-level Stackelberg game: in the inner loop, a MARL agent (the follower) minimizes regret against a fixed population, while in the outer loop an LLM serves as a semantic oracle that generates executable adversarial or cooperative strategies in a Turing-complete code space to maximize the agent's regret. To cope with the absence of gradients in discrete code generation, we introduce Gradient Saliency Feedback, which transforms pixel-level value fluctuations into semantically meaningful causal cues to steer the LLM toward targeted strategy synthesis. We further provide PAC-Bayes guarantees showing that LLM-TOC converges at rate \( O(1/\sqrt{K}) \) and yields a tighter generalization error bound than parameter-space exploration. Experiments on the Melting Pot benchmark demonstrate that LLM-TOC consistently improves zero-shot performance over self-play baselines (IPPO, MAPPO) and the LLM-inference method Hypothetical Minds, while reducing training cost by more than 60%.

Abstract: The rapid digitization of cultural heritage materials has led to the substantial growth of digital library collections, particularly large and heterogeneous archives of periodicals. This expansion has intensified challenges related to content discovery, accessibility, and user engagement, as users increasingly struggle to navigate and identify relevant materials in periodical collections. In this context, intelligent interaction with cultural content has become an essential aspect of effectively accessing and utilizing resources in modern digital libraries, highlighting the need for adaptive and user-oriented mechanisms that support navigation and discovery. Artificial intelligence–driven personalization offers promising solutions to these challenges; however, digital library environments are often characterized by sparse interaction data, evolving user interests, and the continuous introduction of new resources, which limit the effectiveness of standalone content-based or collaborative approaches. This work proposes an integrated personalization approach that combines behavioral interaction data with semantic relationships between documents to support adaptive content delivery in digital libraries. The approach facilitates the discovery of both established and newly digitized or rarely accessed materials, supporting more effective access, exploration, and reuse of large and diverse digital library collections.

Abstract: Recommender systems are increasingly exposed to anomalous user behavior that can distort recommendation outcomes and compromise system reliability. In real-world settings, explicit labels identifying malicious activity are rarely available, motivating the adoption of unsupervised detection approaches. This study presents a comparative analysis of classical machine learning and deep learning techniques for anomaly detection in recommender systems. Using the MovieLens 1M dataset, we construct a user-level behavioral representation based on statistical, temporal, and interaction-based features derived from explicit rating data. Three unsupervised detection models are evaluated: Isolation Forest, One-Class Support Vector Machine, and an autoencoder-based neural network. To address the absence of ground truth labels, evaluation is conducted using label-free protocols, including score distribution analysis, percentile-based thresholding, and inter-model agreement. Results indicate that individual models capture complementary aspects of anomalous behavior, exhibiting low to moderate agreement. An ensemble scoring strategy improves ranking stability and provides a consistent mechanism for identifying highly deviant user profiles. The findings suggest that ensemble-based unsupervised detection constitutes a practical and interpretable first-layer screening approach for recommender system monitoring.

Abstract: This study proposes a Natural Language Processing (NLP)-based evaluation framework to examine the linguistic consistency of Large Language Model (LLM)-generated personalized texts over time. NLP metrics were used to quantify and compare linguistic patterns across repeated generations produced using identical prompts. In Experiment 1, internal reliability was examined across 10 repeated generations from four LLMs (Claude, Llama, Gemini, and ChatGPT) applied to 10 scientific texts tailored for a specific reader profile. Linear mixed-effects models showed no effect of repeated generation on linguistic features (e.g., cohesion, syntactic complexity, lexical sophistication), suggesting short-term consistency across repeatedly generated outputs. Experiment 2 examined linguistic variation across model updates of GPT-4o (October 2024 vs. June 2025) and GPT-4.1 (June 2025). Significant variations were observed across outputs from different model versions. GPT-4o (June 2025) generated more concise but cohesive texts, whereas GPT-4.1 (June 2025) generated outputs that are more academic, lexically sophisticated and complex syntax. Given the rapid evolution of LLMs and the lack of standardized methods for tracking output consistency, the current work demonstrates one of the applications of NLP-based evaluation approaches for monitoring meaningful linguistic shifts across model updates over time.

Abstract: Control charts are widely used in the industrial world to monitor the average and variability of production processes. Max-Half-Mchart is a multivariate control chart that is less effective in handling many outliers. This research aims to develop a control chart that is more resistant to outliers by using Minimum Regularized Covariance Determinant (MRCD). MRCD is a development of the MCD method which is better at dealing with 'fat data', namely situations where the number of variables is greater than the number of observations. The performance evaluation of the robust Max-Half-Mchart control chart based on MRCD using Average Run Length (ARL) against shifts in process mean, process variance, and simultaneous shifts. In addition, a comparison is made of the outlier detection accuracy between the robust Max-Half-Mchart based on MRCD and the standard Max-Half-Mchart. The research results show that the MRCD-based Robust Max-Half-Mchart provides better accuracy and Area Under Curve (AUC) in detecting outliers compared to the traditional Max-Half-Mchart, especially at outlier levels of 10%, 20%, 30%, and 40%. Application of this method to cement quality data also shows superiority in detecting outliers.

Abstract: Large language models exhibit context-dependent behavioral patterns that vary systematically across task domains, yet standardized cross-domain measurement frameworks remain lacking. This study addresses methodological limitations in prior work by applying a rigorous 50-trial protocol uniformly across 14 models (25 model-domain runs) spanning medical (closed-goal) and philosophical (open-goal) reasoning domains using a three-condition protocol (TRUE/COLD/SCRAMBLED). Key findings: (1) domain means show no significant difference (philosophy 0.317 vs medical 0.308; Mann-Whitney U=51, p=0.149), but variance differs markedly (medical SD=0.131 vs philosophy SD=0.045); (2) 23 of 25 model-domain runs show positive ΔRCI, with Gemini Flash medical as the sole negative outlier (ΔRCI=-0.133), suggesting safety filtering interference; (3) vendor signatures show significant differentiation when excluding the Gemini Flash anomaly (F(7,16)=3.55, p=0.017), with Moonshot (Kimi K2) showing highest context sensitivity and Google lowest; (4) the expected information hierarchy (ΔRCI_COLD > ΔRCI_SCRAMBLED) holds in 24/25 runs (96%), validating the measurement framework; (5) position-level analysis reveals prompt-specific variation with a strong P30 summarization spike in medical domain (z=3.74). These results establish ΔRCI as a robust, domain-general metric for context sensitivity and provide the foundation for deeper analyses of temporal dynamics and information-theoretic mechanisms.

Abstract: COVID-19 still poses a global public health challenge, exerting pressure on radiology services. Chest X-ray (CXR) imaging is widely used for respiratory assessment due to its accessibility and cost effectiveness. However its interpretation is often challenging because of subtle radiographic features and inter-observer variability. Although recent deep learning (DL) approaches have shown strong performance in automated CXR classification, their black-box nature limits interpretability. This study proposes an explainable deep learning framework for COVID-19 detection from chest X-ray images. The framework incorporates anatomically guided preprocessing, including lung-region isolation, contrast-limited adaptive histogram equalization (CLAHE), bone suppression, and feature enhancement. A novel four-channel input representation was constructed by combining lung-isolated soft-tissue images with frequency-domain opacity maps, vessel enhancement maps, and texture-based features. Classification was performed using a modified Xception-based convolutional neural network, while Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to provide visual explanations and enhance interpretability. The framework was evaluated on the publicly available COVID-19 Radiography Database, achieving an accuracy of 95.3%, an AUC of 0.983, and a Matthews Correlation Coefficient of approximately 0.83. Threshold optimisation improved sensitivity, reducing missed COVID-19 cases while maintaining high overall performance. Explainability analysis showed that model attention was primarily focused on clinically relevant lung regions.