Human-like Few-Shot Learning via Bayesian Reasoning over Natural Language
This work addresses the problem of bridging the gap between human cognitive efficiency and machine learning models for researchers in cognitive science and AI, though it is incremental in combining existing Bayesian and language model approaches.
The paper tackles the challenge of modeling human-like few-shot concept learning by introducing a Bayesian reasoning model that uses a language model to propose natural language hypotheses, then re-weights them with a prior and likelihood. The result is a model that predicts human judgments across diverse concept types, such as numbers and sets, with demonstrated accuracy.
A core tension in models of concept learning is that the model must carefully balance the tractability of inference against the expressivity of the hypothesis class. Humans, however, can efficiently learn a broad range of concepts. We introduce a model of inductive learning that seeks to be human-like in that sense. It implements a Bayesian reasoning process where a language model first proposes candidate hypotheses expressed in natural language, which are then re-weighed by a prior and a likelihood. By estimating the prior from human data, we can predict human judgments on learning problems involving numbers and sets, spanning concepts that are generative, discriminative, propositional, and higher-order.