Entropy Investigation on Different Critical Properties of Spiking Neurons in Context-Dependent Learning Task

Spiking Neural Networks (SNNs) hold significant importance as a pivotal area of brain-inspired computing. Leveraging an SNN model of the rat hippocampus during the learning of content-related tasks, we examine the trends of coarse-grained entropy concerning the model's key parameters. To quantify this entropy, we adopt a method that maps binary spike sequences into a feature value space. Specifically, for each stimulus condition, the spike sequences of eight neurons are converted into eight corresponding feature values, upon which entropy is calculated by assessing the distribution of these values. We categorize the entropy values based on identical input stimuli and document the frequency of various input stimuli across 130 trials per training session. The entropy of all groups is then summed, weighted by their respective occurrence frequencies. By adjusting the values of the key parameters, we can trace the trends of entropy and derive specific findings. These insights offer invaluable guidance for setting our hyper-parameters.