Abstract: Generative transport models learn to move samples from one distribution to another, for example from a patient's magnetic resonance image to a computed tomography (CT) scan or from noise to an image. Recent ones, such as flow and bridge models, are used widely in medical imaging, and most of these studies report faster sampling or higher image scores than diffusion. In medicine these measures are incomplete; a synthetic CT scan, for example, must also keep the patient's anatomy and support dose calculation. We therefore review 383 studies of flow and bridge models published from 2023 to 2026, identified by a systematic search, with comparison samples of 55 diffusion and 17 normalizing-flow studies, and compare them across clinical tasks on model design and evaluation endpoints. We find that a model's medical value is shaped by the path the patient's information takes to the output, more than by whether the transport is a diffusion, flow or bridge variant. We present these models as settings of one transport viewpoint with six design axes. We then review four task families, reconstruction, translation, generation and analysis, and distil fifteen observations on the information a model receives and on how well its metrics and comparisons support its claims. For each clinical use we identify the endpoint it requires, which conventional metrics do not replace, together with the datasets and sampling costs a fair comparison needs, and give a procedure for choosing and testing a model. A continually updated reading list is maintained at https://github.com/chqwer2/Awesome-Flow-Based-Medical-Imaging.

Abstract: Wind-Assisted Propulsion Systems (WAPS) are moving from pilot projects to commercial deployment, yet no mandatory, internationally accredited training standard exists for the seafarers operating them, and crew preparation relies on non-regulated, manufacturer-led familiarisation. This study assesses operational competency needs based on serving crew members rather than desk-based analysis. A sequential case study combined onboard observation with a structured questionnaire administered to 28 seafarers at a shipping company operating Flettner rotor vessels; the 16 respondents with hands-on rotor experience represent an estimated 5–10% of the global population of such officers. Over 93% of active operators report regular manual intervention or close monitoring, and 56% experience recurring conflicts between their professional judgment and manufacturer-provided system recommendations, challenging the assumption that automation minimises the human element. A lack of holistic understanding of vessel behaviour under aerodynamic forces emerges as the primary obstacle to safe operation, and seafarers prioritise stability and COLREG compliance over efficiency. The study maps the competency areas requiring formalised training and proposes a modular, STCW-aligned curriculum to inform ongoing IMO deliberations on training for wind-propelled ships.

Abstract: Bradyrhizobium exhibits exceptional genetic versatility, functional adaptability, and evolutionary resilience, serving as a critical stabilizing agent in tropical ecosystems. However, its full potential remains underexplored, largely because soil health assessments have historically relied on temperate region and studies often analyze traits in isolation rather than tracking genomic responses to combined environmental stresses. This review applies the niche breadth triangle framework to synthesize the genomic structure, evolutionary history, and multi stress tolerance of tropical Bradyrhizobium lineages. We traced the evolutionary trajectory of the strains, detailing how ancestral lineages fist adapted to to acidic, metal rich soils before acquiring mobile symbiosis islands and specialized secretion systems that broaden their host range. In addition, we highlight that tropical lineages harbor an expansive accessory genome of ~39% variable accessory genes, showcasing a significantly more promising adaptive potential than temperate Bradyrhizobium species. Key protective mechanisms reviewed include modified heat shock promoters, robust extracellular matrices, and photosynthetic metabolism, which preserve nitrogenase activity during drought, high temperatures, and flooding. High strain diversity ensures functional redundancy and niche complementarity, maintaining consistent N-fixation under pressure. Consequently, this work demonstrates how local management practices and stress resistant strains can improve rhizosphere conditions that simultaneously building climate resilient agricultural systems.

Abstract: Technology can substitute for internal cognition, reducing investment through disuse of internal abilities, or complement it by amplifying learning capacity. This paper develops a dynamic model of human capital formation in which individuals allocate cognitive effort between internal reasoning and external technological support. It suggests that which mode predominates depends on the curvature of returns to human capital: diminishing returns induce cognitive offloading, while convex returns generate path-dependent amplification. The main implication is that technological change widens dispersion in cognitive outcomes across individuals and environments rather than simply shifting mean cognitive capacity.

Abstract: Always-on personal intelligence demands more than episodic, task-driven agents. To remain always-on, it requires a continuing loop that sustains assistance between requests, retains personal context, and carries unfinished commitments forward. To grow more personal over time, this loop can also become self-evolving, revising how it works based on past experience. Yet sustaining and improving this loop as a whole remains an open challenge. We therefore organize the assistant into three functional modules, Sensing, Memory, and Action, connected by six directed channels that pass information between them. These modules operate under real-world limits, from battery life to privacy and user permissions, and run on models and the scaffolding around them. We review each module's capabilities, policies, and evaluation methods, and then trace cross-module dependencies across integrated systems to identify bottlenecks. To assess the loop completeness of existing systems, we introduce the Harness for Always-On Loop Oversight (HALO), combining literature and code reviews with executable tests. Our synthesis reveals the thinnest support for two channels: Memory guiding Sensing, and Action outcomes returning to Memory. These gaps motivate two research priorities: keeping sensing open to unexpected change for reliable operation, and attributing outcomes to earlier decisions for self-evolution. By identifying missing interfaces and evidence requirements, this roadmap offers the research community an actionable blueprint for building future always-on personal intelligence systems.

Abstract: Background: HIV-1 drug resistance (HIVDR) threatens the durability of antiretroviral therapy (ART) programmes in sub-Saharan Africa, where non-B subtypes predominate, yet national data linking mutational patterns to patient-level correlates remain scarce. Methods: We analysed 521 Kenyan patients with confirmed virologic failure referred for genotypic resistance testing (2020–2025). Resistance was interpreted with the Stanford HIVdb (v10.1), and pairwise dependencies between high-level resistance classes were quantified descriptively with mutual information (MI) normalised to marginal entropy. Multivariable logistic regression identified factors associated with high-level NRTI resistance in patients with complete covariate data (n = 299). Results: High-level resistance was present in 386 patients (74.1%, 95% CI 70.2–77.7) for NRTIs, 401 (77.0%, 73.2–80.4) for NNRTIs and 179 (34.4%, 30.4–38.5) for PIs; M184V (65.3%) and K103N (34.9%) predominated. NRTI and PI resistance co-accumulated (MI = 0.121 nats, 19% of PI entropy), whereas dependencies involving NNRTI resistance were weak (MI = 0.015, 3%). Older age was the only factor associated with high-level NRTI resistance (adjusted OR 1.05 per year, 95% CI 1.02–1.08); neither subtype nor NRTI backbone was. Conclusions: Extensive multiclass resistance is already established at virologic failure, supporting earlier failure detection, timely regimen switching and expanded access to dolutegravir-based regimens.

Abstract: Post-Acute Sequelae of SARS-CoV-2 (PASC), clinically designated as Long COVID, affects over 400 million individuals globally, presenting as a multisystem disorder characterized by chronic microvascular endotheliitis, exertional exhaustion, pathological cardiac remodeling, and systemic autoimmunity [1–3]. While acute COVID-19 is primarily initiated by viral entry via ACE2, persistent post-viral sequelae are driven by sustained expression of viral accessory factors and host transcriptional hijacking [1,2,4]. Here, we establish a comprehensive systems-biology framework demonstrating that SARS-CoV-2 accessory protein Orf7b overexpression drives high-amplitude transactivation of master oncogene c-Myc, which hypertranscribes long non-coding RNA PVT1 [1,2]. The resultant multi-isoform surge of PVT1 linear, circPVT1, and ecDNA transcripts acts as a high-capacity competing endogenous RNA (ceRNA) network that sequesters protective host microRNA-145-5p (miR-145-5p) and miR-214 [1,2]. This dual microRNA depletion initiates a secondary derepression cascade involving hypoxia-inducible factor-2α (HIF-2α / EPAS1) and Transferrin Receptor 1 (TFRC / TfR1) [1,2,4,5]. Dense TFRC accumulation accelerates diferric transferrin endocytosis, expanding the intracellular labile iron pool, driving ROS-dependent transcription and secretion of chemokine CCL2, and provoking concentric cardiac hypertrophy, microvascular endotheliitis, ACE2-independent viral entry, and TLR7-regulated TFR1-driven B-cell autoimmunity [1–6]. Crucially, we demonstrate that this entire pathological network can be reversed upstream by host-directed p53/NRF2-based epigenetic shielding via miR-145-5p transactivation, 3'-UTR transcript cleavage, and mitochondrial cristae restoration [1,2,7].

Abstract: EEG-based emotion recognition faces challenges from signal non-stationarity, low signal-to-noise ratios, inter-subject variability, and ambiguous emotional boundaries. To address these issues, this study proposes an approach that combines the Multi-Taper Method (MTM) and Welch method for power spectral density (PSD) feature extraction and combines the Multi-Scale Residual Attention and Bidirectional Long Short-Term Memory (MSRA-BiLSTM) for emotion recognition. The approach captures spatial and frequency-band features while learning temporal dependencies. Experiments on three-class classification using the DEAP dataset achieved accuracies of 98.03%, 97.54%, and 96.81% for Valence, Arousal, and Dominance, respectively, with Macro-F1 scores of 0.9789, 0.9749, and 0.9674. The overall accuracy was 97.46%, outperforming conventional Welch-based features, differential entropy features, and end-to-end training. Therefore, the study has contributed to the problem of emotion recognition based on EEG signals.

Abstract: Small impact craters within the Martian north polar layered deposits (NPLDs) provide localized geomorphic records of surface modification within an ice-rich polar environment. We investigate a small crater in Mare Boreum using CTX and multi-temporal HiRISE imagery, HiRISE-derived topography, and regional constraints from MOLA, TES, GRS, SHARAD, and geological mapping. The crater retains a sub-circular planform but exhibits markedly subdued topographic expression, including attenuated rim relief and a shallow interior. HiRISE-derived topography indicates substantially reduced relief relative to fresh-crater morphometric expectations for a feature of comparable scale. The coexistence of a preserved planform and strongly attenuated relief indicates substantial cumulative post-impact modification. Multi-temporal HiRISE observations resolve the crater rim, interior, and illumination-sensitive surface features under broadly comparable seasonal conditions. Regional geological, thermophysical, geochemical, and radar observations constrain the broader ice-rich polar setting but do not resolve crater-scale composition, infill, or subsurface structure. The observed morphology is consistent with cumulative modification that may include depositional infilling, sublimation-related redistribution, rim degradation, and relaxation within the NPLDs, although the relative contributions of these processes cannot be uniquely determined from the available orbital observations.

Abstract: Pentatricopeptide repeat (PPR) proteins constitute one of the largest protein families in plants and play important roles in organellar RNA metabolism, plant growth and development, and stress responses. In strawberry (Fragaria spp.), an economically important crop of the Rosaceae, the composition, evolutionary history and expression patterns of this family have not been characterized systematically. Using the genome assemblies of diploid woodland strawberry (Fragaria vesca) and octoploid cultivated strawberry (Fragaria × ananassa), we identified PPR genes genome-wide with HMMER and integrated subfamily classification, chromosomal mapping, phylogeny, gene duplication, selection pressure and transcriptome expression analyses. A total of 603 and 1785 PPR genes were identified in F. vesca and F. × ananassa, respectively, with P:PLS ratios of 1.29 and 1.27 that agree closely with the value of 1.20 obtained for the positive control Arabidopsis thaliana. Most PPR genes lacked introns, and single-exon genes accounted for 61.5% and 42.3% of the two repertoires. F. vesca PPR genes were unevenly distributed among the seven chromosomes, with the largest number on Fvb6 (121), whereas F. × ananassa PPR genes were distributed among 28 chromosomes (43 to 105 genes per chromosome). Phylogenetic analysis resolved the P and PLS subfamilies into two clearly separated clades and revealed numerous species-specific expansion clusters at the tips of the Fragaria branches. Duplication analysis showed that expansion of the F. vesca PPR family depended mainly on tandem duplications and intra-chromosomal segmental duplications, whereas expansion in F. × ananassa proceeded mainly through inter-chromosomal segmental duplications (retention of subgenome homeologs, 96.4%). Selection pressure analysis showed that 97.1% of duplicated pairs had K_a/K_s < 1 and 71.75% had K_a/K_s < 0.5, indicating strong purifying selection; the mean K_s of F. vesca (1.02) was much higher than that of F. × ananassa (0.116), reflecting differences in the timing of duplication. Transcriptome analysis showed pronounced differential expression of PPR genes among tissues and across fruit developmental stages. Under cold stress, 45 differentially expressed PPR genes were identified in F. vesca (23 down-regulated and 22 up-regulated), whereas no gene passed the stringent threshold under ABA treatment in F. × ananassa and only 29 candidate genes were obtained. This study provides the first systematic characterization of the strawberry PPR gene family and a resource of candidate genes for dissecting PPR function in the Rosaceae and for genetic improvement of stress tolerance and fruit quality in strawberry.

Abstract: Six years after the first licensed mRNA vaccine, the argument in the field is no longer whether RNA can be delivered. It is which RNA, for how long, and at what cost in reactogenicity. Three molecular formats carry most of the work now: nucleoside-modified linear mRNA, self-amplifying RNA (saRNA) and circular RNA (circRNA). Each has produced human or near-human data since 2021, and each is limited at a different step-cytosolic lifetime, innate sensing of replication intermediates, and manufacturing. Beneath all three sits the ionizable lipid, whose discovery process has itself been redesigned: combinatorial libraries, machine learning and iterative chemical evolution have compressed what once took years into months, without yet producing a molecule that separates potency from inflammation. This Review follows the three formats through their mechanistic constraints and their clinical readouts, then examines what the lipid literature of the past five years does and does not establish. We give particular weight to results that conflict, to effect sizes measured in the wrong species, and to claims that circulate without a same-condition comparison behind them.

Abstract: Radio-frequency (RF) antenna design and fabrication traditionally proceeds in three phases. First, the design performance is modeled with expensive, proprietary software. Second, the design is built using precision metal machining or printed circuit board (PCB) fabrication. Third, the result is characterized using RF measurement tools. The process can include machine learning for optimization. We have developed an open-source alternative process that utilizes the MIT Electromagnetic Equation Propagator (MEEP) package for design and simulation, and open-source 3D printing with conductive polymer for fabrication. Using this process, we created a set of broadband RF horn antennas. The radiation patterns and the VSWR match our MEEP calculations, while the cross-polarization ratio matches theoretical expectations. These results indicate that our creations are linearly polarized, broadband RF horn antennas with bandwidth [4.0-18.0] GHz, or a fractional bandwidth of 127%. Because the maximum frequency of our network analyzer is 18 GHz, these results are lower limits. Our results therefore represent a successful broadband open-source CEM/CAD workflow utilizing direct, repeatable conductive printing that circumvents the need for electroplating processes.

Abstract: We prove that the packing chromatic number of the generalized Sierpiński graph over the five-cycle is six in every dimension at least two. Together with the known values for the other cycle bases of length at least four, this completes their classification. Two explicit 125-vertex templates and twenty oriented profile states provide the construction. Each state admits two independently selectable productions, yielding at least 20 · 2^((5^(n−3)−1)/4) optimal labelled colorings in dimension n ≥ 3; their normalized base-two logarithmic count has a limit of at least 1/500. For Sierpiński path products, we determine the maximum packing chromatic number throughout the previously unresolved range m ≥ 12, n ≥ 3: it is seven exactly when m ≥ 20, n ≥ 5, and six otherwise. The proof gives two sharp structural guarantees for trees with at most two pendant paths at each spine vertex: nineteen spine vertices, or sixteen vertices bearing two arms of length at least two, suffice for six colors. Explicit examples show both guarantees are sharp. The finite computations have standalone certificates; the unbounded claims follow from the stated constructions, closure arguments, and induction.

Abstract: This study investigated the implementation of talent management practices among SME owner-managers in South-West Nigeria; with particular attention to their understanding of talent management, practices implemented and their leadership approaches as well as the challenges they faced in implementing talent management practices. The study adopted a quantitative cross-sectional survey design which involved 435 owner-managers and knowledgeable managers of SMEs in Lagos, Ogun, Oyo, Osun, Ondo and Ekiti States. Descriptive and multiple regression analysis was used to analyse the data. The results showed that respondents acknowledged talent management and took measures to identify talent, hire talent, develop talent, motivate and retain talent, as part of the leadership's responsibility. Transformational leadership turned out to be the most significant factor in implementing talent management, whereas financial limitations, lack of HRM knowledge, and operational pressures and challenges in attracting and retaining knowledgeable workers were significant barriers. In conclusion, the study suggests that leadership and proactive employee capability development are crucial to effective talent management. It makes recommendations for SME owner-managers to improve employee development and improve leadership by using participative and transformational leadership and set up practical talent identification and retention strategies. There should also be HRM training and workforce development assistance available to and accessible by government agencies and SME support organisations.

Abstract: This study aimed to establish the prevalence of non-pathogenic Neisseria spp. in urban and rural community in Cameroon and assess the antimicrobial susceptibility patterns. This cross-sectional study collected 629 oral rinse samples; 329 individuals from the urban University of Bamenda community and 300 inhabitants from the rural community of Santa, Cameroon between May to July 2024. A thousand-fold dilution of each sample was plated on selective media for growth of non-pathogenic Neisseria spp. with and without antibiotics. The prevalence of non-pathogenic Neisseria spp. within the study participants was 75.4% (474/629). The high prevalence was also observed within the different study site; 75.6% (249/329) in the University community and 75% (225/300) in Santa community. Non-pathogenic Neisseria species from the University community showed higher resistance rates to azithromycin (75.6%), ciprofloxacin (78.2%), and ceftriaxone (54,4%) than were recorded from Santa community; azithromycin (59.9%), ciprofloxacin (49.1%) and ceftriaxone (33.4%). The high prevalence of antibiotic resistance in non-pathogenic Neisseria spp. underscores the importance of understanding their potential role in spreading antimicrobial resistant genes.

Abstract: Privacy concerns may discourage the use of artificial intelligence in legal services while also motivating interest in platforms with credible safeguards. These responses are often collapsed into a single adoption construct, obscuring whether respondents reject the technology or condition its use on protection. This study examines that distinction in 385 adult, AI-aware survey records from Bangladesh, including 242 law students. Five statements separately measure disclosure concern, self-attributed inhibition, conditional willingness under strong privacy controls, preference for meaningful data control, and the perceived influence of trust. Analyses combine joint response patterns, paired endorsement contrasts, ordinal correlations and exploratory logistic models with robust covariance estimates. Disclosure concern was endorsed by 56.1%, privacy-related inhibition by 75.6%, conditional willingness under strong controls by 72.7%, meaningful-control preference by 53.2%, and the influence of trust by 79.0%. Inhibition and conditional willingness coexisted in 216 records (56.1%). Concern had a small positive ordinal association with conditional willingness (ρ = 0.149; Holm-adjusted p = 0.013), but its adjusted association with binary endorsement was uncertain (odds ratio 1.29; 95% interval 0.98-1.69). None of the four focal adjusted model tests met the 0.05 threshold after Holm correction. The findings identify conditional acceptance within the recorded sample, rather than an observed privacy paradox or an effect on actual adoption. The article contributes a distinction between concern, attributed inhibition and safeguard-contingent willingness. It recommends evaluating concrete privacy controls through tasks that measure informed choices, alongside technical protection. Undocumented recruitment and fieldwork dates constrain population inference and temporal interpretation.

Abstract: Many production AI pipelines call a generative language model for decisions that need no text at all: which queue a ticket belongs to, whether a prompt is a jailbreak attempt, how urgent an email is. The answer arrives after hundreds of milliseconds of token sampling, with a confidence value that is a predicted token rather than a measurement. This paper documents the design, mathematics and measured record of two open alternatives. Laya is a 421M parameter bidirectional decision model that answers typed questions (choice, score, boolean) about a state in one forward pass of roughly 33 ms on a Tesla T4, trained with policy-gradient reinforcement learning whose reward is a strictly proper scoring rule, so that honest probability reporting is the optimum of the objective; it went from plan to public checkpoints and benchmarks in under two days in September 2026. I cover its architecture, training recipe and a compact results record, including an ablation in which a supervised baseline matches the RL objective on single-step questions. The larger share of the paper belongs to Vega, the architecture that followed. Vega freezes an instruction-tuned language model as a reader and trains a dynamical decision engine of 14M to 31M parameters on its hidden states: each option becomes a Gaussian well in a learned 64-dimensional energy landscape, the question drops a particle with an initial position and momentum, a state-space thermostat sets step size and damping, and the answer is read from Boltzmann occupancy along a damped symplectic trajectory. The engine trains in minutes on one GPU, carries conformal sets, abstention and an out-of-distribution flag as fitted artifacts, and in the vega-U form one byte-identical engine serves four readers from 0.8B to 27B, with new readers attached by a closed-form port fit on unlabelled text. All weights, code, harnesses and known failure modes are public.

Abstract: Long-running coding agents rely on context compaction to overcome finite LLM context windows. Existing harnesses typically perform compaction synchronously, placing expensive summarization on the critical path and disrupting prefix-cache reuse. We deep-dive into compaction behavior and costs by analyzing 13.5M GitHub Copilot agentic coding sessions from 3.2M users, comprising 760M LLM calls. Although only 7.8% of sessions trigger compaction, they account for 44.2% of total token consumption. Compaction also causes substantial foreground delay and sharply reduces prefix-cache reuse. To enable systematic evaluation, we first introduce CompactionBench, a benchmark that models software evolution as persistent, multi-turn coding sessions. It enables controlled evaluation of repeated compaction while preserving executable task-level evaluation and reflecting context-growth patterns observed in production. We then present APEC, an asynchronous progressive compaction mechanism that prepares fine-grained, indexed summaries in the background and appends them to a stable prefix, moving compaction work off the critical path while preserving prefix reuse. APEC is harness-agnostic and treats summarization as a black box. Evaluated across frontier models, APEC improves task quality while reducing total cost by up to 25.4% and agent execution time by 9.6%.

Abstract: Predicting cellular responses to genetic and chemical interventions supports disease research and therapeutic discovery. Experimental screening is costly, limiting coverage of interventions and cellular contexts. This survey reviews artificial intelligence virtual cells (AIVC) through the progression from state representation learning to perturbation prediction and intervention design. We describe how state representations support response modeling and how learned responses guide intervention selection across previously untested interventions and diverse cellular contexts. We summarize data resources, representative model families, evaluation strategies, and multimodal, temporal, and spatial extensions, then discuss generalization and experimental feedback. A curated resource collection is available at https://github.com/ttruan2426-dot/Awesome-Single-Cell-Perturbation-Prediction.

Abstract: We present a QRS detector built from a difference of two triangular moving averages, \(x_p=\mathrm{TMA}(x-\mathrm{TMA}(x,w_1),w_2)\), computed by running sums at constant cost per sample, and a local decision rule with no learned parameters. The filter rejects the constant component exactly; we prove a two-sided bound between each triangular stage and the Gaussian of equal variance and a uniform bound, valid for any scale ratio, between the filter and an explicit difference of Gaussians. The decision rule combines a sliding-window threshold with a refractory and T-wave constraint; one of 48 prespecified configurations was selected on the MIT-BIH Arrhythmia Database and tested against Pan--Tompkins and Elgendi's two-moving-average detector on the Supraventricular Arrhythmia, INCART and European ST-T databases (the last never used before) and on the Noise Stress Test benchmark. Elgendi's detector has the highest F1 at \(\pm100\)~ms on every clinical database, but the difference is significant (record-bootstrap 95% interval excluding zero) only on INCART; the proposed detector has a higher gross F1 than Pan-Tompkins on all of them, significantly only on MIT-BIH. Its R-peak timing error has a lighter tail than Pan-Tompkins everywhere and a lighter or equal tail compared with Elgendi's detector (offset-free 95th percentile 22-39~ms against 31-55~ms on three databases; equal on European ST-T). On this benchmark, at 12 and 6~dB, within the band where QRS detection in wearable ECG is considered reliable, it has the highest F1 and fewest false detections (significantly at 6~dB); at 0~dB and below all methods become unusable (\(F_1\le0.33\)).