Frontiers in systems neuroscience · 2026
Exploring the spiking neural autoencoder: from hyperexcitability to noise-driven compensation.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Khodashenas Mohammadreza, Martins Daniel P
Аннотация
INTRODUCTION: Understanding how artificial neural networks (ANNs) can capture biologically meaningful dynamics is a central challenge in systems neuroscience. In this work, we investigate whether spiking neural networks (SNNs) can function not only as machine-learning tools but also as biologically inspired computational analogs and tractable testbeds for studying pathological neural dynamics.
METHODS: We implemented a spiking autoencoder composed of Leaky Integrate-and-Fire and Synaptic neuron models to create a controlled framework for analyzing how biologically related parametric changes to neuronal and synaptic dynamics influence learning and information transfer. By tuning model parameters to induce persistent overfiring-like behavior, we emulated a hyperexcitability-like regime conceptually analogous to NaV channel dysfunction in hippocampal circuits. Reconstruction performance and network activity were evaluated under both noiseless and noisy conditions.
RESULTS: The induced hyperexcitability-like regime degraded image reconstruction performance and disrupted stable information propagation, consistent with impaired processing in hyperexcitable neural systems. Layer-wise firing-rate analysis revealed that the altered regime was characterized by unstable activity redistribution rather than sustained global overactivation. Importantly, introducing controlled Gaussian noise into the input stream partially restored reconstruction quality and improved learning performance, suggesting that stochastic perturbations can partially compensate for instability in dysfunctional network regimes.
DISCUSSION: These findings demonstrate that specific SNN parameter regimes can reproduce key signatures of pathological excitability while also providing a platform for investigating compensatory mechanisms. Overall, this work positions spiking autoencoders as scalable, biologically grounded frameworks for hypothesis-driven studies of neural dysfunction and candidate interventions, supporting the integration of ANN methodologies with mechanistic models in systems neuroscience.
METHODS: We implemented a spiking autoencoder composed of Leaky Integrate-and-Fire and Synaptic neuron models to create a controlled framework for analyzing how biologically related parametric changes to neuronal and synaptic dynamics influence learning and information transfer. By tuning model parameters to induce persistent overfiring-like behavior, we emulated a hyperexcitability-like regime conceptually analogous to NaV channel dysfunction in hippocampal circuits. Reconstruction performance and network activity were evaluated under both noiseless and noisy conditions.
RESULTS: The induced hyperexcitability-like regime degraded image reconstruction performance and disrupted stable information propagation, consistent with impaired processing in hyperexcitable neural systems. Layer-wise firing-rate analysis revealed that the altered regime was characterized by unstable activity redistribution rather than sustained global overactivation. Importantly, introducing controlled Gaussian noise into the input stream partially restored reconstruction quality and improved learning performance, suggesting that stochastic perturbations can partially compensate for instability in dysfunctional network regimes.
DISCUSSION: These findings demonstrate that specific SNN parameter regimes can reproduce key signatures of pathological excitability while also providing a platform for investigating compensatory mechanisms. Overall, this work positions spiking autoencoders as scalable, biologically grounded frameworks for hypothesis-driven studies of neural dysfunction and candidate interventions, supporting the integration of ANN methodologies with mechanistic models in systems neuroscience.