Encoding Sensory and Motor Patterns as Time-Invariant Trajectories in Recurrent Neural Networks

TL;DR

This paper demonstrates how recurrent neural networks can encode, recognize, and transcribe complex time-varying patterns like spoken digits by forming stable neural trajectories that generalize across speed and spatial variations.

Contribution

It shows that tuning RNN weights enables encoding of temporal patterns as invariant trajectories, explaining neural mechanisms for temporal and spatial generalization.

Findings

RNNs can recognize spoken digits by encoding them as neural trajectories.

Neural trajectories exhibit temporal invariance through modulated angular velocities.

The model predicts mechanisms for neural generalization across time and space.

Abstract

Much of the information the brain processes and stores is temporal in nature - a spoken word or a handwritten signature, for example, is defined by how it unfolds in time. However, it remains unclear how neural circuits encode complex time-varying patterns. We show that by tuning the weights of a recurrent neural network (RNN), it can recognize and then transcribe spoken digits. The model elucidates how neural dynamics in cortical networks may resolve three fundamental challenges: first, encode multiple time-varying sensory and motor patterns as stable neural trajectories; second, generalize across relevant spatial features; third, identify the same stimuli played at different speeds - we show that this temporal invariance emerges because the recurrent dynamics generate neural trajectories with appropriately modulated angular velocities. Together our results generate testable predictions as to how recurrent networks may use different mechanisms to generalize across the relevant spatial and temporal features of complex time-varying stimuli.