Input Driven Synchronization of Chaotic Neural Networks with Analyticaly Determined Conditional Lyapunov Exponents

TL;DR

This paper explores how external inputs can synchronize chaotic recurrent neural networks by analyzing the stability through conditional Lyapunov exponents, providing analytical and numerical tools for understanding synchronization.

Contribution

It introduces an analytical method to determine the stability of synchronized states in chaotic RNNs using conditional Lyapunov exponents and master stability functions.

Findings

Synchronization is achievable when the weight matrix has zero row-sum.

The stability depends on the largest real eigenvalue of the weight matrix.

Conditional Lyapunov exponents can be computed analytically for certain signals.

Abstract

Recurrent neural networks (RNNs) with random, but sufficiently strong and balanced coupling display a well known high-dimensional chaotic dynamics. Here, we investigate if externally applied inputs to these RNNs can stabilize globally synchronous, input-dependent solutions, in spite of the strong chaos-inducing coupling. We find that when the balance between excitation and inhibition is exact, that is when the row-sum of the weights is constant and 0, a globally applied input can readily synchronize all neurons onto a synchronous solution. The stability of the synchronous solution is analytically explored in this work with a master stability function. For any synchronous solution to the network dynamics, the conditional Lyapunov spectrum can be readily determined, with the stability of the synchronous solution critically dependent on the largest real eigenvalue component of the RNN weight matrix. We find that the smaller the maximum real component of the weight matrix eigenvalues, the more readily the network synchronizes. Further, the conditional Lyapunov exponents are easily computed numerically for any synchronization signal without simulating the RNN. Finally, for certain oscillatory synchronization signals, the conditional Lyapunov exponents can be determined analytically.