Quality of internal representation shapes learning performance in feedback neural networks

Lee Susman,Francesca Mastrogiuseppe,Naama Brenner,Omri Barak

DOI: https://doi.org/10.1103/PhysRevResearch.3.013176

2020-11-12

Abstract:A fundamental feature of complex biological systems is the ability to form feedback interactions with their environment. A prominent model for studying such interactions is reservoir computing, where learning acts on low-dimensional bottlenecks. Despite the simplicity of this learning scheme, the factors contributing to or hindering the success of training in reservoir networks are in general not well understood. In this work, we study non-linear feedback networks trained to generate a sinusoidal signal, and analyze how learning performance is shaped by the interplay between internal network dynamics and target properties. By performing exact mathematical analysis of linearized networks, we predict that learning performance is maximized when the target is characterized by an optimal, intermediate frequency which monotonically decreases with the strength of the internal reservoir connectivity. At the optimal frequency, the reservoir representation of the target signal is high-dimensional, de-synchronized, and thus maximally robust to noise. We show that our predictions successfully capture the qualitative behaviour of performance in non-linear networks. Moreover, we find that the relationship between internal representations and performance can be further exploited in trained non-linear networks to explain behaviours which do not have a linear counterpart. Our results indicate that a major determinant of learning success is the quality of the internal representation of the target, which in turn is shaped by an interplay between parameters controlling the internal network and those defining the task.

Neurons and Cognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to understand how the learning performance in feedback neural networks is affected by the interaction between the internal network dynamics and the characteristics of the target signal. Specifically, the researchers focused on the learning performance of nonlinear feedback networks when generating sinusoidal signals. By precisely mathematically analyzing the linearized network model, they predicted that the learning performance is optimal when the target signal has a specific intermediate frequency. This frequency monotonically decreases as the internal reservoir connection strength increases. At the optimal frequency, the reservoir's representation of the target signal is high - dimensional and asynchronous, and thus has the maximum robustness to noise. In addition, the study also found that this relationship can be further utilized to explain the behaviors in the trained nonlinear networks, which have no counterparts in the linear model. Overall, the research results show that the main determinant of learning success is the quality of the internal representation of the target signal, which in turn is shaped by the interaction between the parameters controlling the internal network and the parameters defining the task.