Latent Event-Predictive Encodings through Counterfactual Regularization
This addresses the challenge of segmenting sensorimotor information into compact event encodings for intelligent systems, with potential applications in hierarchical reasoning, planning, and decision making, but it appears incremental as it builds on existing theories of event-predictive cognition.
The paper tackled the problem of inferring structure from continuous data streams by introducing a SUrprise-GAted Recurrent neural network (SUGAR) with counterfactual regularization, which learned to compress temporal dynamics into latent event-predictive encodings and anticipate event transitions on a hierarchical sequence prediction task.
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.