Latent Event-Predictive Encodings through Counterfactual Regularization

TL;DR

This paper introduces SUGAR, a neural network model with counterfactual regularization that learns to encode and predict event sequences, enabling better understanding and anticipation of environmental dynamics in complex data streams.

Contribution

The paper presents a novel neural network architecture, SUGAR, that uses counterfactual regularization to produce compositional, event-predictive encodings for hierarchical sequence prediction.

Findings

Model effectively compresses temporal dynamics into latent codes.

Enables accurate anticipation of event transitions.

Latent codes exhibit compositional properties.

Abstract

A critical challenge for any intelligent system is to infer structure from continuous data streams. Theories of event-predictive cognition suggest that the brain segments sensorimotor information into compact event encodings, which are used to anticipate and interpret environmental dynamics. Here, we introduce a SUrprise-GAted Recurrent neural network (SUGAR) using a novel form of counterfactual regularization. We test the model on a hierarchical sequence prediction task, where sequences are generated by alternating hidden graph structures. Our model learns to both compress the temporal dynamics of the task into latent event-predictive encodings and anticipate event transitions at the right moments, given noisy hidden signals about them. The addition of the counterfactual regularization term ensures fluid transitions from one latent code to the next, whereby the resulting latent codes exhibit compositional properties. The implemented mechanisms offer a host of useful applications in other domains, including hierarchical reasoning, planning, and decision making.