Continuous-Time Neural Networks Can Stably Memorize Random Spike Trains

Area of Science:

• Computational Neuroscience
• Artificial Neural Networks

Background:

• Recurrent neural networks (RNNs) are foundational in processing sequential data.
• Precisely timing neural activity is crucial for information processing in biological systems.

Purpose of the Study:

• To investigate the capacity of continuous-time RNNs for storing and recalling exact spike train timings.
• To assess the robustness and accuracy of temporal recall in these networks.

Main Methods:

• Numerical experiments were conducted on continuous-time RNN models.
• Synaptic weights were computed offline based on a template promoting temporal stability.
• The network's ability to memorize and reproduce spike trains was evaluated.

Main Results:

• Continuous-time RNNs can robustly memorize random spike train scores across all neurons.
• Networks accurately reproduce spike timings with high probability within a specific parameter range.
• Associative recall was demonstrated even under noisy input conditions.

Conclusions:

• Continuous-time RNNs offer a viable mechanism for precise temporal information storage and recall.
• The findings support the potential of these networks in modeling neural computation involving exact spike timing.
• The demonstrated temporal stability and associative recall capabilities are significant for future applications.