2026-03-24
Han He, Ruinan Chen, Yixiang Wang, Xia Chen
The ability to anticipate future events continuously is a hallmark of biological vision, yet standard deep learning models often struggle with long-term coherence due to the rigid discretization of time. In this paper, we propose NeuralVisionNet, a probabilistic framework that models visual anticipation as a continuous generative process, drawing inspiration from the predictive coding mechanisms of the hippocampal-entorhinal circuit. Our architecture synergizes hierarchical Video Swin Transformers with Attentive Neural Processes, employing a novel grid-like coding scheme to represent spatiotemporal dynamics as a continuous function rather than a fixed sequence of frames. Furthermore, we introduce a variational global latent variable to encode the “event gist,” ensuring semantic consistency over extended horizons. Extensive evaluations on KTH, Human 3.6M, and UCF 101 benchmarks demonstrate that NeuralVisionNet significantly outperforms state-of-the-art stochastic baselines in perceptual quality (FVD) and structural fidelity (SSIM), offering a robust computational proof-of-concept for continuous, bio-inspired visual forecasting.
DOI: 10.3389/fncom.2026.17810802026-03-18
Yunming Zhang
IntroductionIn classroom scenarios, student behaviors exhibit high intra-class variance and subtle inter-class differences, while complex backgrounds and severe occlusions pose significant challenges for accurate behavior recognition.MethodsSBR-YOLO is proposed as a student behavior detection framework for accurate and robust recognition in complex classroom environments. To address the challenges posed by visually similar behaviors and non-uniform spatial distributions of targets, a Behavior-aware Context-Position Attention module is designed, which leverages learnable positional encoding and inter-head interaction mechanisms to capture spatial dependencies among behavioral regions and enable discriminative feature learning. To handle substantial scale variations between front-row and back-row students, an Adaptive Spatial Feature Fusion mechanism is introduced at each output level of the neck, prior to the detection heads, which adaptively learns fusion weights for cross-scale feature integration. A Class-Aware Discriminative Loss function is further introduced to enhance fine-grained discrimination by enforcing intra-class compactness and inter-class separation constraints.ResultsExperiments on SCB-Dataset3 demonstrate that SBR-YOLO achieves 74.2% mAP@50, representing a 6.4 percentage point improvement over the YOLOv8n baseline, with the parameter count increasing moderately from 3.0 M to 4.6 M.DiscussionComprehensive ablation studies and comparative experiments with state-of-the-art methods confirm the effectiveness of SBR-YOLO for student behavior recognition in complex smart classroom environments.
DOI: 10.3389/fncom.2026.18044222026-02-13
Martin N. P. Nilsson
IntroductionThis paper presents a comprehensive mechanistic model of a neuron with plasticity that explains how information input as time-varying signals is processed and stored. Additionally, the model addresses two long-standing, specific biological challenges: Integrating Hebbian and homeostatic plasticity, and identifying a concise synaptic learning rule.MethodA biologically accurate small-signal equivalent-circuit model is derived through a one-to-one mapping from established ion-channel properties. The often-overlooked dynamics of the synaptic cleft is essential in this process. Analysis of the model reveals a simple and succinct learning rule, indicating that the neuron functions as an internal-feedback adaptive filter, a common concept in signal processing.ResultsSimulations confirm the model's functionality, stability, and convergence, demonstrating that even a single neuron without external feedback can act as a potent signal processor. The model replicates several key characteristics typical of biological neurons, which are seldom captured in other neuron models. It can encode time-varying functions, learn without risking instability, and bootstrap from a state where all synaptic weights are zero.DiscussionThis paper explores the function of neurons with a focus on biological accuracy, not computational efficiency. Unlike neuromorphic models, it does not aim to design devices. The electronic circuit analogy aids understanding by leveraging decades of electronics expertise but is not intended for physical implementation. This interdisciplinary work spans a broad range of subjects within the realm of neurobiophysics, including neurobiology, electronics, and signal processing.
DOI: 10.3389/fncom.2026.17165592026-02-13
Haoming Yang, Marko Angjelichinoski, Suya Wu, Joy Putney, Simon Sponberg, Vahid Tarokh
Subject-to-subject variability is a common challenge in generalizing neural data models across subjects, discriminating subject-specific and inter-subject features in large neural datasets, and engineering neural interfaces with subject-specific tuning. While many methods exist that map one subject to another, it remains challenging to combine many subjects in a computationally efficient manner, especially with highly non-linear features such as populations of spiking neurons or motor units. Consider subjects with trained neural decoders as sources and those without as targets. Our objective is to transfer data from one or more target subjects to the domain of the source subjects to directly apply the source neural decoder such that no target decoder needs to be trained. We propose to use the Restricted Boltzmann Machine (RBM) with Gaussian inputs and Bernoulli hidden units; once trained over the entire feature set of subjects, the RBM allows the mapping of target features on source feature spaces using Gibbs sampling. We also consider a novel computationally efficient training technique for RBMs based on the Fisher divergence, which allows closed-form gradients of the RBM to be computed. We apply our methods to decode turning behaviors from neuromuscular recordings of spike trains from the ten muscles that primarily control wing motion in an agile flying hawk moth, Manduca sexta. The dataset consists of this comprehensive motor program recorded from nine subjects, each driven by six discrete visual stimuli. The evaluations show that the target features can be decoded using the source classifier to classify the visual stimuli with an accuracy of up to 95% when mapped using an RBM trained by Fisher divergence, suggesting that RBMs for multi-cross-subject mapping applications are effective and efficient.
DOI: 10.3389/fncom.2026.17109142026-02-13
Natasha Clarke, Abigail E. Licata, Soumaiya Imarraine, Thuy Dao, Ginevra Sperandio, Ana Luísa Pinho, Valentina Borghesani, Paola Mengotti, Antonietta Gabriella Liuzzi, Doris Pischedda
IntroductionDespite the increasing representation of women in scientific fields, disparities in research funding allocation remain. This inequity deprives talented women researchers of necessary resources, limiting the diversity of perspectives and ideas, and contributes to the “scissor-shaped curve” seen in neuroscience, where women leave before obtaining senior positions. Data transparency and comprehensive reporting of information on grant winners and applicants, as well as reporting of gender and other intersecting demographics and key metrics, are crucial to effectively evaluate funding equity. However, there is a lack of guidelines on which data funders should report. In this study, we aimed to investigate the transparency of neuroscience funders across Europe, focusing on the European Union, Schengen area, and the United Kingdom.MethodsTo this end, we developed a Transparent Reporting Scale (TRS), composed of 15 items crucial to facilitate transparent and meaningful reporting, and searched for public data from funders in order to apply the scale and evaluate their transparency in data reporting. Across 32 countries and the European Union as a whole, we identified 39 funders, with 90% sharing publicly available data on funding results.ResultsUsing the TRS, five funders received a “gold” rating, eighteen a “silver” one, and thirteen a “bronze” rating. Scale scores were significantly correlated with the Gender Equality Index [p = 0.64, 95% CI (0.33, 0.83), p = 0.001] and gross domestic product of the countries where funders are based [p = 0.51, 95% CI (0.20, 0.74), p = 0.003], suggesting that collection and/or publication of funding data may reflect overall commitments to gender equity, and be limited due to resources. Data from only 29% of funders could be disaggregated for the neuroscience category specifically, indicating the difficulty in evaluating equity in our field.DiscussionWe collated all available data into an Open Science Framework repository to enable data sharing and further analyses. The TRS can support funders in adopting transparent, standardized reporting practices in order to support evidence-based progress toward gender equity.
DOI: 10.3389/fncom.2026.17652492026-02-11
Ki Yun Lee, M. Taher A. Saif
Mechanical forces have recently emerged as critical modulators of neural communication, yet their role in high-level cognitive functions remains poorly understood. Here, we present a biologically inspired spiking neural network model that integrates mechanical tension, vesicle dynamics, and spike-timing-dependent plasticity to examine how tension influences learning, memory, and cognitive operations such as pattern completion, projection, and association. We find that increased tension enhances synaptic efficiency by accelerating vesicle clustering and recovery, resulting in a 67% improvement in memory recall speed and a 17% increase in inter-regional synchrony during projection relative to relaxed states. Conversely, a 20% reduction in tension leads to a 31% decline in memory association performance, highlighting the tension-sensitive accessibility of stored information. The model further reveals that an appropriate balance of inhibition is essential for these tension-driven effects: networks with 20% inhibitory neurons achieve optimal spatial precision in memory encoding and recall, whereas insufficient inhibition allows tension-amplified excitation to spread uncontrollably and degrade recall fidelity. Together, these in silico findings position mechanical tension as a functional neuromodulator and suggest new directions for neuromorphic design and energy-efficient, living computing platforms.
DOI: 10.3389/fncom.2026.1737434