Oscillatory dynamics in complex recurrent neural networks

TL;DR

This paper demonstrates how oscillatory activity can emerge from a biologically inspired recurrent neural network model, linking complex-valued inputs to oscillations observed in brain activity.

Contribution

It introduces a mathematical framework showing complex-valued inputs lead to oscillations in recurrent neural networks, validated through simulations.

Findings

Complex-valued inputs are equivalent to combined real-valued networks.

Recurrent neural networks can produce oscillatory signatures.

Simulation confirms the emergence of oscillations from the proposed model.

Abstract

Spontaneous oscillations measured by Local field potentials (LFPs), electroencephalograms and magnetoencephalograms exhibits variety of oscillations spanning frequency band ($1-100$ Hz) in animals and humans. Both instantaneous power and phase of these ongoing oscillations have commonly been observed to correlate with pre-stimulus processing in animals and humans. However, despite of numerous attempts it is not fully clear whether the same mechanisms can give rise to a range of oscillations as observed in vivo during resting state spontaneous oscillatory activity of the brain. In the current paper we show how oscillatory activity can arise out of general recurrent on-center off-surround neural network. The current work shows (a) a complex valued input to a class of biologically inspired recurrent neural networks can be shown to be mathematically equivalent to a combination of real-valued recurrent network with real-valued feed forward network, (b) such a network can give rise to oscillatory signatures. We also validate the conjecture with results of simulation of complex valued additive recurrent neural network.