A Self-Supervised Framework for Function Learning and Extrapolation
This work addresses the problem of few-shot generalization and extrapolation for machine learning and biological agents, representing an incremental advance by building on existing function learning paradigms.
The paper tackles the challenge of enabling agents to generalize and extrapolate in high-dimensional environments by proposing a self-supervised framework that acquires general-purpose representations for function learning. The result shows that the framework outperforms other models in downstream tasks, including extrapolation, though specific numerical gains are not provided.
Understanding how agents learn to generalize -- and, in particular, to extrapolate -- in high-dimensional, naturalistic environments remains a challenge for both machine learning and the study of biological agents. One approach to this has been the use of function learning paradigms, which allow peoples' empirical patterns of generalization for smooth scalar functions to be described precisely. However, to date, such work has not succeeded in identifying mechanisms that acquire the kinds of general purpose representations over which function learning can operate to exhibit the patterns of generalization observed in human empirical studies. Here, we present a framework for how a learner may acquire such representations, that then support generalization -- and extrapolation in particular -- in a few-shot fashion. Taking inspiration from a classic theory of visual processing, we construct a self-supervised encoder that implements the basic inductive bias of invariance under topological distortions. We show the resulting representations outperform those from other models for unsupervised time series learning in several downstream function learning tasks, including extrapolation.