搜索结果：Neurocomputing

This study develops a bioinspired bilayer volatile memristor with a W/SiO₂/Cux(SiO2)100-x/Cu structure. The morphology and crystalline structure of the conductive filaments were directly observed via cross-sectional transmission electron microscopy. Discrete spherical copper-based grains construct conductive filaments with large internal surfaces, some of which are distorted due to stress interactions with the silicon dioxide matrix. According to the results of x-ray photoelectron spectroscopy, the diffusion of copper and redox reactions (involving the valence state transitions of copper to zero valence, positive monovalence, and positive divalence) are the core mechanisms governing the dynamic evolution of conductive filaments. When the stimulus is subsequently removed, the minimization of the thermodynamic surface energy drives the transformation of nonspherical grains into stable spherical grains, leading to conductive filament rupture and spontaneous recovery of the device to the initial state. By regulating the parameters of pulse signals to achieve precise control over conductive filament dynamics, the device successfully reproduces the behavior of nociceptors. A high accuracy of 93.11% for the handwritten digit recognition task in neurocomputing is achieved, showing the multipurpose function of the memristor.

The present study designs a new computational approach to study the thermal performance of a shrinking or stretching longitudinal rectangular fin under both convective and radiative conditions. A dimensionless mathematical model is developed to describe the thermal behavior of the fin, including parameters such as Peclet number, convective and radiative coefficients, temperature ratios, and stretching/shrinking parameters. The hybrid neurocomputing method used to solve the nonlinear system of differential equations consists of Cascaded Neural Networks (CNN) along with Genetic Algorithm hybridized with Sequential Quadratic Programming (GA-SQP). CNN modeling offers an approximate solution through a piecewise continuous representation, whereas GA-SQP improves convergence by optimal network weights. A detailed parametric study is done in six scenarios to assess the fin's temperature distribution and tip characteristics. The accuracy and stability of the solver are validated using statistical metrics such as RMSE, MAE, E-VAF, and E-NSE. The outcome shows that the Peclet number increases enhance the fin tip temperature up-to 15% as the temperature ratio increases, the fin tip temperature increases by 7-10%. The increase in the the radiation coefficient reduces the temperature by 8%. Various statistical measures confirm its reliability and effectiveness, making it a strong tool for handling complex ODEs and PDEs. Histogram and boxplot analysis show accuracy ranges up to 10- 9, indicating that almost 90% of total iterations achieved the required accuracy tolerance range. Fitness evaluation attains high rate of convergence shows that CNN-GA-SQP scheme is a fast-converging and accurate computational intelligence approach for solving stiff nonlinear fin heat transfer models.

Language switching exemplifies a real-world model of adaptive control, as bilinguals select languages in response to continuously changing contextual demands. Although the Adaptive Control Hypothesis (ACH) highlights context-dependent language control in bilinguals, how these strategies are learned remains unclear. Drawing on reinforcement learning (RL) theory, we examined whether reward prediction error (RPE) drives adjustments in voluntary language switching. Chinese-English bilinguals performed a voluntary picture-naming task with probabilistic (i.e., high, medium, and low) reward feedback that generated RPE depending on the switch decisions. Computational modeling revealed that RPE dynamically updated an abstract, generalizable switch policy. Representational similarity analysis (RSA) suggested the learned value of the control policy was represented in the MTL, whereas RPE representations emerged in the dorsolateral prefrontal cortex (dlPFC). Furthermore, connectome-based predictive modeling (CPM) revealed a multi-stage network process. A distributed, cross-network pattern initially supported early exploratory learning, which then transitioned into a centralized, hub-centric network for policy exploitation. Finally, the network specialized into a localized circuit for policy automation, coupled with a globally distributed network for monitoring unexpected errors. Together, these findings establish adaptive language control as a value-based RL process. This provides a neurocomputational framework for strategy learning, extending RL principles from simple stimulus-response to high-level cognitive control.

The blood oxygen level-dependent (BOLD) signal has been instrumental in characterizing brain activity. While the spatial resolution of fMRI continues to improve, relatively few methods have focused on enhancing and leveraging temporal resolution to investigate the spatiotemporal dynamics of the hemodynamic response. In this study, we applied a reordering method to achieve ultra-high temporal resolution (60 ms) in data acquired during a somatosensory stimulation paradigm. We then used a finite impulse response model (FIR) for each participant (N = 31) to preserve the temporal dynamics in the statistical analysis. At the group level, we employed an ANOVA combined with 4D nonparametric permutation testing to identify significant signal changes in time across the whole brain. Our results characterize the hemodynamic response in terms of both its temporal and spatial patterns and reveal distinct differences in response shapes within the somatosensory system. This method introduces a time-resolved approach to BOLD signal analysis, drawing inspiration from grand-average techniques commonly used in EEG research.

Social and communicative deficits are defining characteristics of autism spectrum disorder (ASD). Some theories suggest that these challenges, among other autistic traits, may arise from differences in predictive coding, or how the brain uses context to predict and interpret incoming information. This idea has the potential to link symptoms of autism to specific neurocomputational processes, and is especially promising for communication, whose impairment is a hallmark of ASD. Here we leveraged the ability of large language models (LLMs) to quantify semantic contextualization to analyze a unique dataset of responses from hippocampal neurons obtained during language listening in three mild-to-severe autistic individuals with comorbid epilepsy. Key elements of semantic coding were preserved in all three individuals with ASD: single-neuron response dynamics, representation of word-word semantic relationships, and patterns of context-dependent shifts in meaning. However, relative to controls, ASD resulted in reduced neural signatures of contextualization: (1) neuronal responses were aligned with earlier, less contextual layers of GPT-2, (2) ASD patients had lower effective dimensionality of the neural subspace predicting semantics, (3) neural representations of word meaning were less influenced by preceding context, and (4) neural signatures of lexical surprisal were reduced. Together, these results support theories of autism that emphasize impairments in contextualization, and highlight the power of LLMs as a tool for quantifying the computational basis of neurodevelopmental disorders.

This paper focuses on possible time-domain neurocomputational mechanisms for short-term anticipatory processes. Here we present a simple, signal processing functional model of how short-term rhythmic pattern expectancies could be computed on the fly using recurrent neural timing nets (RTNs). The model is inspired by Gestaltist grouping principles for repeating temporal patterns of events (beats, pulses, grooves, metrical and non-metrical patterns). Building on previous autocorrelation models of pitch, meter, and rhythm, the RTN rhythm perception model consists of temporal codes, temporal pattern memory traces circulating in delay loops, and neural delay-and-coincidence networks with dynamically-adapting spike-correlation-dependent synapses. The network tracks in parallel all event periodicities in rhythmic hierarchies. As in memory trace theories of mismatch negativity (MMN-like) neural responses, it generates simple and complex pattern expectancies and registers deviations from them. Similarities and differences of this correlation-based model with those based on oscillators and predictive coding are discussed.

Predicting the success of startup companies, defined as achieving an exit through acquisition or IPO, is a critical problem in entrepreneurship and innovation research. Datasets such as Crunchbase provide both structured information (e.g., funding rounds, industries, and investor networks) and unstructured text (e.g., company descriptions), but effectively leveraging such heterogeneous data for prediction remains challenging. Traditional machine learning approaches often rely only on structured features and achieve moderate accuracy, while large language models (LLMs) offer strong reasoning capabilities but are not readily adapted to domain-specific business data. We present CrunchLLM, a domain-adapted and backbone-agnostic LLM framework for startup success prediction. CrunchLLM integrates structured company attributes with unstructured textual narratives and applies parameter-efficient fine-tuning together with prompt optimization to specialize foundation models for entrepreneurship data. Importantly, our framework introduces a self-verifiable multitask objective, in which the justification loss serves as a training-time constraint on classification, together with a hierarchically ordered input encoding that reduces the tendency of long unstructured company narratives to overshadow structured business attributes. These methodological innovations yield more reliable and feature-grounded predictions than conventional prompt-based LLM adaptation. Our approach achieves 89% accuracy on the Crunchbase startup success prediction task, significantly outperforming traditional classifiers and baseline LLMs. Beyond predictive performance, CrunchLLM generates interpretable reasoning traces that support its predictions, enhancing transparency and trustworthiness for financial and policy decision-makers. Overall, this work demonstrates how domain-aware LLM adaptation and structured-unstructured data fusion can advance predictive modeling of entrepreneurial outcomes, providing both a methodological framework and a practical tool for data-driven decision-making in venture capital and innovation policy.

Indirect evidence from preclinical and neuropharmacological studies suggests that oxytocin may attenuate social dominance relationships and modulate social memory. In humans, oxytocin may also modulate self- versus other-oriented reward learning and in-group/out-group decisions differently. Although the neural bases of learning social hierarchy by observation have been identified for linear hierarchies (a1 < a2 < an), the brain mechanisms engaged in learning more complex rank relationships (i.e. <, =, >) in social networks remain unknown. Here, we investigated the modulatory role of oxytocin on the neurocomputational mechanisms engaged when learning ranks in social networks and when making transitive inferences based on social memory retrieval. Participants learned rank relationships between members of two social networks, one in which they were embedded and the other not. During a subsequent test phase, they inferred rank relationships between pairs of members not encountered before. During training, oxytocin (vs. placebo) improved choice accuracy regarding rank relationship between pairs of members for other-oriented network and increased activity of the ventromedial prefrontal cortex. During the test phase, oxytocin modulated the balance between self/other social memory, boosting performance and amygdala responses for memory retrieval concerning social networks in which participants were not embedded. These findings demonstrate that oxytocin modulates the brain computations needed to learn and retrieve rank relationships in social networks in a manner that depends upon reference-frame.

Sensory errors, mismatches between predicted sensory outcomes of movement and reafferent sensory feedback, drive changes in the feedforward control of future motor behavior that correct for those errors. Across a wide variety of motor behaviors, individuals with cerebellar damage show impairments in these corrections, strongly suggesting a key role of the cerebellum in sensorimotor adaptation. However, the extent to which the cerebellum is involved in controlling vocal pitch is currently unknown. Crucially, vocal pitch differs in several ways from other systems that suggest it relies more on feedback than feedforward control. Adaptation itself also differs in vocal pitch: Rather than the gradual buildup/decay of learning seen in other systems, pitch adaptation and deadaptation are almost immediate. Together, this questions whether adaptation in vocal pitch relies on the same mechanism as other motor domains. Here, we test the hypothesis that the cerebellum underlies sensorimotor adaptation in vocal pitch, testing the domain generality of this neurocomputational process. In both sustained vocalization and a more natural word production task, individuals with cerebellar ataxia fail to adapt to external auditory perturbation of vocal pitch. The lack of adaptation observed, compared to the impaired but present adaptation seen in other systems, suggests that the cerebellum plays an especially critical role in maintaining accurate control of vocal pitch. Conversely, we failed to detect a previously observed increase in online compensation to vocal pitch errors in ataxia, potentially suggesting this may be an idiosyncratic change in control rather than a common trait in this population.

Urdu is spoken by over 230 million people worldwide, yet it remains significantly underrepresented in digital resources, with limited availability of large-scale, publicly accessible training datasets for optical character recognition (OCR). The diversity of Urdu font styles encountered in printed books, newspapers, and digital publications poses a substantial barrier to developing generalizable OCR systems, while the absence of standardized benchmarks hinders fair and reproducible comparison across recognition approaches. This data article presents FIPU-OCR-CHAR, a benchmark dataset of printed Urdu characters encompassing 48 classes: 38 alphabets and 10 numerals in their isolated forms. The dataset was constructed through a fully systematic pipeline comprising five sequential stages: font collection and validation, character set definition, base image rendering, augmentation, and dataset organization with split generation. Each character class was rendered from 201 distinct Urdu TrueType/OpenType font files, producing 9,648 base images (201 fonts &#xd7; 48 classes). Each base image was subsequently processed through 34 augmentation operations encompassing geometric transforms, photometric adjustments, blur filters, noise injection, and morphological operations, producing 328,032 augmented images. The complete dataset totals 337,680 labeled PNG images, each stored at 28&#xd7;28 pixel resolution with 24-bit color depth. The dataset is organized into three predefined splits: training (70%; 236,376 images), validation (20%; 67,536 images), and testing (10%; 33,768 images), each accompanied by a CSV annotation file mapping image filenames to integer class labels (0-47). The repository additionally contains a Jupyter Notebook implementing a ResNet-34 baseline classification pipeline, a results summary image, and a README file documenting dataset structure and label definitions. The dataset is publicly available on Mendeley Data under a CC BY 4.0 license and is intended for use in OCR model development, font-invariant classifier training, Urdu script digitization, transfer learning for word- and line-level recognition, and benchmarking of convolutional neural network and Vision Transformer architectures on low-resource script character recognition tasks.

The overlapping symptoms between bipolar disorder (BD) and major depressive disorder (MDD) pose a challenge in diagnosis and treatment. A prevailing hypothesis suggests that mood dysregulation may be linked to impairments in the reward system, but the neurocomputational differences between BD and MDD remain elusive. This study investigates whether atypical reward processing affects subjective mood in adolescents with BD and MDD. Our research aims to elucidate the behavioral and neural differences between the two groups, facilitating more accurate and timely diagnosis and intervention. Forty-five adolescents (aged&#xa0;&#x2264;&#xa0;19 years) diagnosed with BD-II in depressed mood states (N&#xa0;=&#xa0;25) or MDD (N&#xa0;=&#xa0;20) completed a risky gambling task while their brain responses were recorded using functional magnetic resonance imaging (fMRI). Several computational models were constructed to uncover the associations between various reward components (e.g. reward prediction errors, RPE) and trial-wise fluctuations in subjective mood during the task. Adolescents with BD exhibited a lower best choice rate and a higher uncertain choice rate compared to those with MDD. Computational modeling and mediation analysis suggested a tripartite mediating relationship between RPE-mood association, decision rationality, and symptom severity. Using fMRI, we observed significant RPE-related activation in the ventral striatum, which showed a slight positive correlation with the RPE-mood association. We also noted subtle differences in several brain regions (i.e. medial orbitofrontal cortex) between the BD and MDD groups. These differences were further associated with manic symptoms. Decision rationality mediated the association between RPE-mood association and symptom severity. Relative to adolescents with MDD, those with BD showed decreased decision rationality, along with modest but distinct reward-related neural patterns on fMRI. These findings highlight the crucial role of reward processing in mood regulation and provide preliminary neurocomputational evidence that may inform future diagnostic biomarker development.

We introduce a brain-constrained neurocomputational model designed to simulate higher cognitive functions of the human brain, implemented using NEST, a widely used open-source simulator optimised for high-performance spiking neural network simulations. Previously implemented in the custom-built C-based Felix simulation library, transitioning the model to NEST enhances accessibility, reproducibility, and computational efficiency. At the cellular level, the model comprises spiking excitatory neurons and local inhibitory neurons, whereas at the network level, it replicates the structural and functional organisation of 12 cortical regions spanning frontal, temporal, and occipital cortices, along with their associated inter-area connectivity. Additionally, global inhibition mechanisms and neuronal noise are integrated. Learning in the model follows biologically plausible Hebbian plasticity principles, incorporating both long-term potentiation and long-term depression. To validate the NEST implementation, we replicated previous simulation findings obtained with the Felix-based model. The new implementation successfully reproduced the same topographical distribution of cell assemblies following associative learning of object and action words within action and perception systems, replicating a range of previous neuroimaging results. Although the NEST model produced larger cell assemblies than Felix, the overall topographical patterns remained similar, indicating preservation of fundamental network characteristics. Moreover, the transition to NEST significantly enhanced computational efficiency, reducing simulation runtime nearly sixfold compared to Felix. This improvement in computational speed is crucial for expanding the model to include additional cortical regions, such as extending to the right hemisphere, which necessitates increased computational resources. The online version contains supplementary material available at 10.1007/s11571-026-10415-5.

Background: The global prevalence of Alzheimer's disease (AD) has reached 55.2 million. AD is characterized by progressive deterioration in cognition and working memory (WM), which are essential for attention, reasoning, and learning. These impairments are associated with pathological changes in cortical and subcortical regions. Binaural beats (BBs), a non-invasive auditory neuromodulation technique, have demonstrated cognitive enhancement effects in healthy individuals; however, their impact on WM in patients with AD remains largely unexplored. Methods: This study investigated the effects of BB stimulation on WM and cognitive function in the temporal lobe of patients with AD using standardized Low-Resolution Electromagnetic Tomography (sLORETA). Twenty-five patients with AD were randomly assigned to either an experimental group (n = 15) that received BB stimulation or a control group (n = 10) that received standard auditory stimulation. EEG recordings were obtained before and after the intervention. Results: Paired t-tests conducted on timeframe and frequency-wise sLORETA images revealed significant increases (p < 0.05) in theta, alpha1, and alpha2 frequency bands in the experimental group. Activated regions included the inferior, middle, superior, and transverse temporal gyri; Brodmann areas (BA) 20, 21, 22, 40, and 42; as well as networks associated with working memory and cognition. Conclusions: These findings suggest that BB stimulation induces temporal lobe activation, thereby enhancing working memory and cognitive function in patients with AD.

The neuroscience of planning has long been analogized to search algorithms in artificial intelligence (AI), which simulate future actions to guide immediate choices. We argue that advances in both neuroscience and AI suggest that planning is better understood to encompass a broader class of computations where mental simulation supports learning, often well before a decision is needed. We review three neurocomputational mechanisms that illustrate this shift. First, hippocampal replay resembles search but also often occurs prospectively or offline, likely training downstream circuits rather than directly guiding choice. Second, temporally abstract representations, such as grid cells, can enable planning without iterative search. Third, metalearning may shape how prefrontal dynamics implement task-specific planning strategies, echoing how AI systems learn to adapt across contexts. This view recasts the brain's planning machinery as a family of learning processes that leverage simulations to build representations and strategies, with forward search as one special case.

Objective.Microstimulation delivers electrical pulses directly into the brain, with one of its promises being to restore lost senses to millions of people. Yet a fundamental challenge remains: how do intracortical microstimulation (ICMS) patterns engage neural circuits to achieve the inception of specific experiences, such as vivid sensory percepts of touch and vision? Here, we define 'inception' as the initiation of percepts evoked by microstimulation through the mapping of stimulation to circuit-level activity that results in sensory experiences.Approach.This perspective proposes an integrated research framework that combines Reverse Translation, Forward Translation, and computational neuroscience to bridge insights between clinical observations and high-resolution animal studies.Framework.Our framework envisions the development and evaluation of ICMS strategies within a cross-species system that narrows the range of plausible underlying neural mechanisms and the set of evoked perceptual outcomes. Reverse Translation uses human perceptual reports about phosphenes, tones, and touch to guide investigations in rodents and non-human primates, mapping the cell types and circuits underlying each percept. Forward Translation leverages these biological insights to design refined ICMS protocols for selective circuit engagement. Bidirectional Translation weaves these approaches together through computational neuroscience, ensuring that experimental observations iteratively and continuously refine one another across species and experimental modalities.Significance.This integrated strategy aims to transform microstimulation research into a dynamic dialogue between fundamental science and human experience. Harnessing the Bidirectional Translation Framework can accelerate therapies that enhance quality of life for people with sensory or motor impairments, and contribute more broadly to systems neuroscience by uncovering the mechanisms by which causal manipulation changes activity in neurons and networks.

暂无摘要（点击查看详情）

Vestibular dysfunction is a common cause of dizziness and a leading cause of medical visits. Yet, current assessment methods of dizziness remain largely subjective, relying on self-reports and intermittent clinical evaluations that lack real-time monitoring, quantitative precision, and preventive capability. This paper introduces EquilibriSense, a bio-inspired, head-worn system for quantifying motion-induced dizziness under a controlled head-rotation paradigm. The system integrates multiple physiological sensing modalities with an AI-driven pipeline and a neurocomputational labeling framework to model dizziness progression and classify symptom severity on a five-level scale (0-4). In a pilot study involving a small cohort of 10 healthy participants in a controlled laboratory setting, the system achieved 86.8% accuracy in multi-level motion-induced dizziness classification and enabled early detection of dizziness onset with an AUC of 0.99 and over 98% accuracy, supported by high precision and recall. These results demonstrate the feasibility of using multimodal physiological sensing to characterize motion-induced dizziness and establish EquilibriSense as a proof-of-concept platform for objective dizziness quantification, providing a foundation for future validation in real-world and clinical populations.

Previous research has shown that favorable monetary outcomes relative to others enhance neural responses; however, these paradigms often confound the subjective utility of social comparison with absolute reward magnitude. It remains unclear how performance-based social comparisons modulate an individual's self-efficacy and the subsequent effort mobilization. To address this gap, we conducted behavioral (N&#x2009;=&#x2009;32) and electrophysiological (N&#x2009;=&#x2009;34) experiments using a self-paced effortful task with downward/upward social comparison feedback. Computational modeling and multilevel mediation analyses revealed that downward (vs. upward) feedback dynamically updated self-efficacy, which fully mediated subsequent effort mobilization. Electrophysiological results showed stronger neural responses to downward versus upward feedback. Critically, downward comparison enhanced preparatory markers for the next round, including the contingent negative variation (CNV) and cue-beta power. Moreover, higher self-efficacy predicted more negative-going CNV potentials. These findings demonstrate that performance-based comparisons dynamically regulate self-efficacy, shaping both neural preparation and effort allocation in goal-directed behavior.

Major depressive disorder (MDD), bipolar disorder (BP), and schizophrenia (SCZ) involve learning impairments with poorly understood mechanisms. Understanding both the similarities and differences in these mechanisms is important to guide the development of new, targeted interventions. A total of 255 participants diagnosed with MDD (n = 54), BP (n = 47), or SCZ (n = 67) or without any clinical diagnoses (control [CTRL]) (n = 87) performed an associative learning task. Computational modeling quantified the mechanistic interplay between working memory (WM) and reinforcement learning (RL). The latent RL and WM signatures in the electroencephalography (EEG) dynamics showed shared and distinct neurocognitive mechanisms underlying learning. All clinical groups showed learning impairments at the behavioral level. Model-based EEG analyses linked these impairments to distinct patterns in the dynamic interplay between latent RL and WM mechanisms, contrasting with the typical patterns observed in the CTRL group. SCZ was characterized by reduced neural markers of WM, weakening the cooperative influence of WM onto RL (reduced WM recruitment), and reduced integration of negative feedback. Conversely, MDD was characterized by reduced reciprocal influence of RL onto WM, reducing the tendency to upregulate WM contribution with reward history (impaired WM management). Finally, BP was characterized by deficits in both WM and RL recruitment, along with higher WM decay. Behavioral learning impairments that seem similar across clinical groups can be linked to distinct neurocognitive mechanisms via integrative neurocomputational modeling. Our approach provides insights into the interplay of underlying learning mechanisms and how they manifest differently across psychopathologies. People with depression, bipolar disorder, and schizophrenia often show learning difficulties but the underlying causes may differ. By combining brain activity recordings with computational models, we identified distinct cognitive mechanisms driving these impairments. Our findings show how modeling and physiology can give insights into hidden decision dynamics behind learning difficulties. We also outline key steps needed to advance computational psychiatry tools toward clinical applications, including their potential use in guiding personalized treatment.