Frontiers in Computational Neuroscience最新文献

{"title":"Editorial: Health data science and AI in neuroscience & psychology.","authors":"Lei Lu, Zhiqiang Huo, Jin Li, Erying Zhao","doi":"10.3389/fncom.2026.2002192","DOIUrl":"https://doi.org/10.3389/fncom.2026.2002192","url":null,"abstract":"","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"2002192"},"PeriodicalIF":3.3,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13677912/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149223473","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An <i>in silico</i> cortico-basal ganglia-thalamic testbed for comparing visual and deep brain stimulation in Parkinson's disease.","authors":"Chun-Hui Huang, Wen-Shuo Chen, Jin-Rong Lai, Chia-Hung Cho, Yun-Ta Sung, Chih-Tsung Chang, Chii-Wann Lin","doi":"10.3389/fncom.2026.1950516","DOIUrl":"10.3389/fncom.2026.1950516","url":null,"abstract":"<p><strong>Introduction: </strong>Deep brain stimulation (DBS) is an established intervention for Parkinson's disease (PD), whereas the network-level effects of prescribed periodic visual input implemented with 10 Hz base pulses remain uncertain.</p><p><strong>Methods: </strong>We developed a conductance-based cortico-basal ganglia-thalamic (CBT) model comprising 11 Hodgkin-Huxley neuronal populations and a prescribed periodic visual-input unit (RET) to compare 130 Hz subthalamic nucleus (STN) DBS with the archived periodic visual-input condition on the same calibrated PD network. Primary conditions used <i>N</i> = 10 independent runs each. Additional analyses included an unpaired 0-5 mA/cm² DBS-amplitude series, a matched-seed internal-globus-pallidus-to-thalamocortical-relay (GPi→TC) sensitivity analysis at 0.8, 1.0, and 1.2 × nominal coupling (<i>N</i> = 10 per scale), and a matched-seed three-parameter ±20% local sensitivity analysis (<i>N</i> = 5 per parameter-condition cell).</p><p><strong>Results: </strong>DBS reduced the model-internal STN synaptic-current beta measure from 0.0184 ± 0.0060 to 0.0018 ± 0.0004 a.u. (Holm-corrected <i>p</i> = 1.1 × 10^-5), whereas periodic visual input increased thalamocortical relay (TC) and reticular thalamic (RE) firing and enhanced cortical pyramidal (PY) gamma (30-80 Hz) and narrowband 38-42 Hz power. Cortical PY firing was 4.86 ± 0.84 Hz in PD, 5.28 ± 0.87 Hz with DBS (<i>p</i> = 0.285 versus PD), and 10.33 ± 0.85 Hz with periodic visual input (Holm-corrected <i>p</i> = 1.18 × 10^-10 versus PD). The ampltude series showed a non-monotonic STN response at 1 mA/cm²; from 2 to 5 mA/cm², mean STN beta values occupied a narrow range of 0.00186-0.00212 a.u. In the GPi→TC analysis, TC spontaneous firing remained 0 Hz in 30/30 runs, and all 750,000 saved voltage samples were finite and subthreshold (-72.06 to -54.74 mV; no 0-mV crossings). Across the three-parameter analysis, DBS raw STN-beta suppression remained 88.9%-95.8%, and periodic visual input increased TC firing 3.70-4.29-fold relative to nominal PD.</p><p><strong>Discussion: </strong>These findings show persistence across the tested dose, GPi→TC, and three-parameter local domains but do not establish evoked relay fidelity, global parameter robustness, biological mechanisms, or clinical efficacy.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1950516"},"PeriodicalIF":3.3,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13672486/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149215378","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A sliding window and Fourier decomposition method based 1D-CNN framework for motor imagery EEG classification.","authors":"B S S Tejesh, Karra Bhaskar, M Sivaji, Ravisankar Dakupati, Saswat Kumar Ram, Pradyut Kumar Sanki, M Ramakrishnan","doi":"10.3389/fncom.2026.1945160","DOIUrl":"10.3389/fncom.2026.1945160","url":null,"abstract":"<p><p>Electroencephalography (EEG) signal classification is fundamental requirement for effective motor imagery (MI) brain-computer interface (BCI) applications. However, current MI classification methods often depend on handcrafted spatial filtering or complex time-frequency transformations that increase computational complexity. To mitigate this limitation, we propose a unified architecture that integrates a sliding window (SW), a Fourier decomposition method (FDM), and a one-dimensional convolutional neural network (1D-CNN) for MI-EEG classification. The raw EEG data were segmented using the SW approach to augment the data and extract more temporal feature patterns from each trial. The segmented signal, containing time-domain features, was subsequently processed using a band-pass filter and FDM to extract frequency-based features. Here, FDM decomposes the signal into eight orthogonal Fourier Intrinsic Band Functions (FIBFs). A 1D-CNN architecture was employed to learn representations from the decomposed EEG signals. The proposed framework was evaluated on the BCI Competition III Dataset IVa (BCIC III IVa) and BCI Competition IV-2a (BCIC IV 2a) datasets. Our method achieved accuracies of 79.14% and 93.67% on the BCIC IV 2a and BCIC III IVa datasets, respectively, under the subject-dependent evaluation protocol. To evaluate cross-subject generalization, the proposed framework was validated using the Leave-One-Subject-Out (LOSO) protocol, achieving average classification accuracies of 76.49% on the BCIC III Dataset IVa and 55.50% on the BCIC IV 2a dataset. Furthermore, deployment on the NVIDIA Jetson Nano demonstrated complete pipeline latencies of 58.119 ms/trial and 85.055 ms/trial, corresponding to throughputs of 17.21 and 11.76 trials/s for the BCIC III IVa and BCIC IV 2a datasets, respectively. The results show the strong potential of the developed MI-EEG classification method to assist subjects with motor ailments.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1945160"},"PeriodicalIF":3.3,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13672498/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149215358","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MEMOIR-VLM-a multimodal vision-language model for Alzheimer's disease classification and question answering.","authors":"Sohail Haresh Gidwani, Tamoghna Chattopadhyay, Sophia I Thomopoulos, Paul M Thompson","doi":"10.3389/fncom.2026.1902258","DOIUrl":"10.3389/fncom.2026.1902258","url":null,"abstract":"<p><strong>Introduction: </strong>Alzheimer's disease (AD) affects an estimated 55 million people worldwide and is projected to nearly double by 2050, creating a need for scalable, non-invasive AI systems that can classify disease stage, support multimodal clinical reasoning, and interact in natural language. Most deep learning work on AD focuses on single-modality classification without natural-language interaction and typically assumes complete input data.</p><p><strong>Methods: </strong>We present MEMOIR-VLM (Multimodal Encoder with Missing-modality and Open-ended Inference for Retrieval), a modular two-stage vision-language framework trained and evaluated on 2,363 ADNI subjects. The first stage is a missing-modality-aware encoder that fuses T1-weighted MRI and DTI fractional anisotropy maps with structured clinical scores via cross-attention fusion with stochastic modality dropout, allowing a single set of weights to operate on any of the seven non-empty modality subsets without imputation. After CLIP/InfoNCE contrastive pretraining, the encoder is fine-tuned for multi-task prediction of 3-way diagnosis, binary diagnosis, CDR-SB, age, and sex. The second stage extends the frozen encoder into a retrieval-augmented generation pipeline for case-based reasoning and open-ended visual question answering.</p><p><strong>Results: </strong>On held-out ADNI test data, with CDR-SB excluded from the inputs to remove a cognitive-score leak, the encoder achieved 91.3% balanced accuracy for CN vs. dementia and 68.2% balanced accuracy for 3-class diagnosis (macro-F1 = 0.681); age was predicted with a 5.96-year MAE. When the diagnosis field was withheld from retrieved captions, no evaluated LLM exceeded a retrieval-only k-NN majority vote (0.673), which itself closely matched the encoder's 3-class accuracy (0.682). The RAG-VQA layer showed strong text fidelity (BERTScore = 0.894; SBERT cosine similarity = 0.811). Zero-shot external validation on OASIS-3 (1,048 subjects, no diffusion data) retained 78.7% balanced accuracy for CN vs. impaired (AUC = 0.889) using identical ADNI-trained weights with the DTI branch masked.</p><p><strong>Discussion: </strong>MEMOIR-VLM combines missing-modality-aware multimodal encoding, multi-task clinical prediction, and retrieval-grounded natural-language reasoning within a single modular framework. The results support using the encoder, optionally with k-NN voting, as the diagnostic component and the RAG-VQA layer as an interpretable natural-language interface that surfaces comparable cases and readable clinical summaries. The external validation results further demonstrate the practical value of the missing-modality mechanism when an imaging modality is unavailable.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1902258"},"PeriodicalIF":3.3,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13668829/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149215429","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Network-level restoration of impaired memory function in early Alzheimer's disease via GluN2B modulation: insights from a hippocampal CA1-CA3 model.","authors":"Fabio Librizzi, Saana Seppälä, Marja-Leena Linne, Michele Migliore, Ausra Saudargiene","doi":"10.3389/fncom.2026.1953352","DOIUrl":"10.3389/fncom.2026.1953352","url":null,"abstract":"<p><p>Memory storage and recall are essential tasks for the healthy function of the hippocampus, adversely affected by neurodegenerative diseases, such as Alzheimer's Disease (AD). Pathological levels of amyloid-β peptide (Aβ), found already in early-stage AD, can disrupt the NMDAR-dependent plasticity at hippocampal networks. Here we investigate these effects in a biophysically detailed computational model of a hippocampal CA1-CA3 network and examine potential mechanisms of functional recovery. We show that, at a network level, long-term potentiation (LTP) impairments caused by Aβ accumulation can be substantially restored by partially blocking the GluN2B-NMDA receptors-a mechanism previously observed and modeled only at a single cell level. Experimental access to network-level processes of memory storage and recall in pathological conditions remains highly challenging. Therefore, computational modeling may help to bridge the gap between cellular impairments and system-level cognitive dysfunction. Our work transfers the established GluN2B-dependent rescue mechanisms from single cells to a network-level system of hippocampal pyramidal neurons and interneurons. The results demonstrate that also at a higher cognition level, the partial inhibition of the activity of GluN2B-NMDA receptors can be an effective strategy to counteract early AD impairments on memory storage and recall.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1953352"},"PeriodicalIF":3.3,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13668818/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149215434","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An innovative framework based on GMV-ALFF multi-modal MRI and Gaussian data augmentation for psychiatric disorder classification.","authors":"Wenjing Zhu, Xiaoying Jiang","doi":"10.3389/fncom.2026.1835559","DOIUrl":"https://doi.org/10.3389/fncom.2026.1835559","url":null,"abstract":"<p><strong>Objective: </strong>Schizophrenia (SZ) and bipolar disorder (BD) exhibit significant overlap in clinical symptoms, posing substantial challenges to accurate diagnosis. This study proposes an innovative multi-modal fusion framework based on convolutional neural networks (CNNs), integrating structural and functional MRI biomarkers-gray matter volume (GMV) and amplitude of low-frequency fluctuation (ALFF).</p><p><strong>Methods: </strong>Total 31 patients with schizophrenia, 48 patients with bipolar disorder, and 38 healthy controls (HC) were enrolled and underwent resting-state fMRI scans. The framework comprises three core components: Gaussian data augmentation to mitigate the limitation of small sample sizes; a dedicated CNN architecture for processing GMV and ALFF features to generate class probability outputs for SZ, BD, and HC; and DempsterShafer (D-S) evidence theory for robust multi-modal decision fusion.</p><p><strong>Results: </strong>Extensive 5-fold cross-validation confirms the framework's efficacy: it achieves superior performance metrics (Accuracy: 0.9656 ± 0.0324, Precision: 0.9698 ± 0.0319, Recall: 0.9656 ± 0.0324, F1-Score: 0.9661 ± 0.0324) and outperforms traditional machine learning methods (SVM, Random Forest, XGBoost, KNN).</p><p><strong>Conclusion: </strong>The findings demonstrate that the proposed multi-modal fusion approach enhances the ability to differentiate among SZ, BD, and HC compared to unimodal classification methods, providing a reliable technical support for the auxiliary diagnosis of psychiatric disorders.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1835559"},"PeriodicalIF":3.3,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13661931/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149192321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A computational roadmap to activity-dependent myelination.","authors":"Thomas Nguyen, Jeremie Lefebvre","doi":"10.3389/fncom.2026.1895171","DOIUrl":"https://doi.org/10.3389/fncom.2026.1895171","url":null,"abstract":"<p><p>There is growing evidence to support myelin plasticity as a complementary mechanism to synaptic plasticity in the brain. Activity-dependent myelination (ADM) enables oligodendrocytes to modify myelin structure in response to neural activity, modulating axonal conduction velocity and delays to maintain precise signal timing. This introduces a timing-based plasticity mechanism that operates in parallel to changes in signal strength. Despite its significance, ADM lacks a strong theoretical framework due to the interaction of multiple mechanisms operating on widely different time scales. Computational models are beginning to address this gap by providing powerful tools to interpret and guide experimental studies. This perspective explores the growing number of computational ADM models, comparing how they link neural activity to myelin adaptation, and their assumptions, strengths, and limitations. We also discuss the challenges of modeling ADM and open questions, positioning ADM as a key component of neural plasticity.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1895171"},"PeriodicalIF":3.3,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13650009/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149167853","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A high-precision synchronization and centroid-based framework for single-trial decoding of perceived natural speech.","authors":"Cristian Y Olivares, Norelli Schettini","doi":"10.3389/fncom.2026.1947069","DOIUrl":"10.3389/fncom.2026.1947069","url":null,"abstract":"<p><p>Decoding natural speech from EEG is constrained by low signal-to-noise ratios, commonly addressed through event-related potential (ERP) grand averaging. However, this approach acts as a low-pass filter, obscuring trial-to-trial variability and high-frequency neural dynamics. To overcome these limitations, we propose an open-box, FPGA-synchronized framework for single-trial EEG analysis of Spanish vowels (/a/, /i/, /u/). The proposed architecture enforces sub-millisecond hardware determinism, eliminating software-induced temporal jitter and enabling robust extraction of high-frequency oscillatory features, including the Gamma band, without relying on averaged ERPs. This synchronization framework facilitates the separation of early exogenous sensory responses from late endogenous cognitive processing, while comprehensive ablation studies were performed to evaluate the effect of software-induced jitter on decoding performance. Results showed that introducing simulated jitter degraded classification performance to theoretical chance levels (33.20%), confirming hardware synchronization as a strict prerequisite for reliable single-trial decoding. The framework was further validated using a Centroid-PCA-SVM pipeline operating on a 16-channel spatial subset. Using the full proposed architecture, a grand mean classification accuracy of 67.70% was achieved, substantially exceeding the theoretical chance level of 33.33%. The across-participant mean of the maximum fold accuracies reached 83.00%, with individual validation folds achieving up to 90.00%. In contrast, zero-calibration leave-one-subject-out (LOSO) generalization remained at chance level (33.08%; κ_pooled = -0.0036), indicating limited cross-participant transferability and highlighting the current need for participant-specific calibration or alignment. Overall, these findings demonstrate that hardware-deterministic EEG acquisition enables the recovery of robust trial-by-trial brain dynamics that are typically obscured by conventional averaging approaches. By combining precise synchronization with low-latency algorithmic inference, the proposed framework establishes a foundation for real-time neural decoding and represents a stepping stone for practical brain-computer interfaces and advanced clinical monitoring applications.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1947069"},"PeriodicalIF":3.3,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13642302/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149126962","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"How to train your neuron: developing a detailed, up-to-date, multipurpose model of hippocampal CA1 pyramidal cells.","authors":"Luca Tar, Sára Sáray, Máté Mohácsi, Tamás F Freund, Szabolcs Káli","doi":"10.3389/fncom.2026.1897214","DOIUrl":"10.3389/fncom.2026.1897214","url":null,"abstract":"<p><p>Anatomically and biophysically detailed models of neurons have been widely used to study information processing in these cells. Most studies focused on understanding specific phenomena, while more general models that aim to capture various cellular processes simultaneously remain rare even though such models are required to predict neuronal behavior under more complex, natural conditions. In this study, we aimed to develop a detailed, data-driven, general-purpose biophysical model of hippocampal CA1 pyramidal neurons. We leveraged extensive morphological, biophysical and physiological data available for this cell type, and established a systematic workflow for model construction and validation that relies on our recently developed software tools. The model is based on a high-quality morphological reconstruction and includes a diverse curated set of ion channel models. After incorporating the available constraints on the distribution of ion channels, the remaining free parameters were optimized using the Neuroptimus tool to fit a variety of electrophysiological features extracted from somatic whole-cell recordings. Validation using HippoUnit confirmed the model's ability to replicate key electrophysiological features, including somatic voltage responses to current input, the attenuation of synaptic potentials and backpropagating action potentials, and nonlinear synaptic integration in oblique dendrites. Our model also included active dendritic spines, modeled either explicitly or by merging their biophysical mechanisms into those of the parent dendrite. We found that many aspects of neuronal behavior were unaffected by the level of detail in modeling spines, but modeling nonlinear synaptic integration accurately required the explicit modeling of spines. Our data-driven model of CA1 pyramidal cells matching diverse experimental constraints is a general tool for the investigation of the activity and plasticity of these cells and can also be a reliable component of detailed models of the hippocampal network. Our systematic approach to building and validating multi-purpose models should apply to other cell types as well.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1897214"},"PeriodicalIF":3.3,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13638379/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149106224","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Locality and weight sharing shape Hebbian principal-component learning in biologically constrained visual models.","authors":"Patrick Inoue, Florian Röhrbein, Andreas Knoblauch","doi":"10.3389/fncom.2026.1943064","DOIUrl":"https://doi.org/10.3389/fncom.2026.1943064","url":null,"abstract":"<p><strong>Introduction: </strong>Biologically motivated artificial visual learning requires local plasticity rules to be evaluated within architectures whose connectivity assumptions remain explicit. Convolutional Hebbian networks demonstrate that competitive local learning can support visual classification, but their performance is coupled to exact weight sharing, a prior that is not anatomically literal. This study asks whether Hebbian principal component analysis (HPCA) can support representations when spatial locality is preserved but exact translational weight sharing is removed.</p><p><strong>Methods: </strong>HPCA was evaluated on CIFAR-10 as the primary benchmark, with MNIST and STL-10 used as controls. We compared fully connected HPCA, position-specific locally connected HPCA with fixed receptive fields, non-shared synaptic weights, and population-wise divisive response normalization, and HPCA substitutions in two shared-kernel convolutional Hebbian benchmarks. Representations were learned without supervised feature-layer targets or BP-based feature-layer fine-tuning, frozen, and evaluated through supervised readouts.</p><p><strong>Results: </strong>Position-specific locally connected HPCA outperformed the fully connected reference. On CIFAR-10, the best locally connected model reached 61.42 ± 0.36%. Shared-kernel HPCA reached 74.51 ± 0.38% in the modular benchmark with common HPCA preprocessing and 76.57 ± 0.55% in the SoftHebb architecture with reference preprocessing.</p><p><strong>Discussion: </strong>The results show that HPCA remains effective across position-specific locally connected and shared-kernel convolutional architectures, but that performance depends strongly on the surrounding architecture and benchmark protocol. The tested shared-kernel configurations achieved higher accuracy, yet useful frozen visual representations still emerged when exact parameter sharing was removed. Locally connected HPCA therefore provides a controlled model for examining how local plasticity interacts with receptive-field structure and convolutional inductive bias in biologically constrained visual learning.</p>","PeriodicalId":12363,"journal":{"name":"Frontiers in Computational Neuroscience","volume":"20 ","pages":"1943064"},"PeriodicalIF":3.3,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13635126/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149106322","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}