Exploring the spiking neural autoencoder: from hyperexcitability to noise-driven compensation.

Area of Science:

• Systems Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Artificial neural networks (ANNs) offer insights into biological neural dynamics.
• Spiking neural networks (SNNs) serve as biologically inspired computational models.
• Studying pathological neural dynamics requires tractable testbeds.

Purpose of the Study:

• Investigate SNNs as computational analogs for pathological neural dynamics.
• Analyze the impact of biologically related parametric changes on learning and information transfer.
• Emulate hyperexcitability-like regimes to study neural dysfunction.

Main Methods:

• Implemented a spiking autoencoder using Leaky Integrate-and-Fire and Synaptic neuron models.
• Tuned parameters to induce hyperexcitability-like behavior, mimicking NaV channel dysfunction.
• Evaluated reconstruction performance and network activity under noiseless and noisy conditions.

Main Results:

• Hyperexcitability degraded image reconstruction and information propagation.
• Network activity showed unstable redistribution rather than global overactivation.
• Controlled Gaussian noise partially restored reconstruction quality and learning.

Conclusions:

• SNN parameter regimes can replicate pathological excitability signatures.
• Spiking autoencoders provide a framework for studying neural dysfunction and interventions.
• This work integrates ANN methodologies with mechanistic models in systems neuroscience.