ePoster, COSYNE 2023
Synaptic-type-specific clustering optimizes the computational capabilities of balanced recurrent networks
Emmanouil Giannakakis, Anna Levina, Victor Buendia and Sina Khajehabdollahi
University of Tubingen; Department of Computer Science; University of Tubingen; Computer Science; University of Tubingen
- Montreal, Canada
- Computational Neuroscience, Neuroscience and
Abstract
The specific topology of biological neural networks is believed to be one of the most important ingredients of the nervous system's computational capabilities. Furthermore, multiple experimental studies have demonstrated that neural connectivity is highly inhomogeneous, following different patterns between different regions and neuron types. In particular, the presence of clusters of highly interconnected neurons is one of the most well-established connectivity patterns across multiple brain areas. Still, the functional implications of such brain connectivity features, especially for the performance of specific tasks remain largely unknown. Here, we use a reservoir computing model to link diverse, synaptic type-specific levels of clustering in balanced recurrent networks with the ability to process multiple uncorrelated complex inputs simultaneously.
We construct a balanced recurrent network of rate neurons, which we use as a reservoir that simultaneously predicts the trajectory of two chaotic attractors. Predicting the dynamics of a single chaotic system is a commonly used task. For the more challenging simultaneous prediction, we split the network into two clusters, each responsible for processing one attractor. We investigate whether synaptic connections between the two clusters can benefit the reservoir's performance. In particular, we hypothesize that different probabilities of connection within and between clusters for different types of synapses could improve the network's ability to process complex dynamics. Thus, we use approximate Bayesian computation to estimate the distribution of synapse type-specific clustering levels that lead to optimal network performance. We find that localized excitation, combined with more spread-out inhibition (a pattern observed in cortical networks), boosts the network's performance.
Our findings suggest that contrary to common intuition, overlapping connectivity between sub-networks performing different tasks can lead to beneficial dynamics that enhance task performance. Thus, commonly observed connectivity patterns in the brain could have a functional role in the parallel processing of multiple signals.
Topics
- biological neural networks, search World Wide
- neural connectivity, search World Wide
- balanced recurrent networks, search World Wide
- reservoir computing, search World Wide
- rate neurons, search World Wide
- chaotic attractors, search World Wide
- approximate Bayesian computation, search World Wide
- localized excitation, search World Wide
- spread-out inhibition, search World Wide
Details
- Session
- Poster Session I