Background

Epilepsy and other neurological disorders are often associated with abnormal patterns of brain activity that can affect learning, memory, and information processing. Studying these mechanisms directly in the human brain is challenging, so researchers increasingly use computational models to investigate how changes in neural activity influence network function. Spiking Neural Networks (SNNs), which mimic the timing-based communication of biological neurons, provide a promising platform for exploring these questions.

Research

In this study, we developed a spiking neural network autoencoder and introduced a controlled hyperexcitability-like condition inspired by abnormal neuronal activity observed in epilepsy. We examined how this altered state affected the network’s ability to learn and reconstruct information and then investigated whether introducing controlled Gaussian noise could influence performance. Surprisingly, the added noise partially restored network function and improved learning, suggesting that carefully applied perturbations can help stabilize dysfunctional neural dynamics.

Potential Impact

Our findings demonstrate how biologically inspired artificial neural networks can be used as experimental testbeds to study neurological dysfunction and potential compensatory mechanisms. While the work does not directly model epilepsy or provide a clinical treatment, it offers a framework for generating hypotheses about how abnormal neural activity affects information processing and how targeted interventions might help restore function. In the longer term, such computational approaches may contribute to the development of new strategies for understanding and treating disorders characterized by disrupted brain network dynamics.