Human-like Few-Shot Learning via Bayesian Reasoning over Natural Language

TL;DR

This paper presents a human-like few-shot learning model that uses Bayesian reasoning over natural language hypotheses, effectively predicting human judgments across diverse concept types.

Contribution

It introduces a novel Bayesian framework that leverages language models and human data to improve concept learning in a human-like manner.

Findings

Accurately predicts human judgments on concept learning tasks

Handles diverse concept types including generative and propositional

Demonstrates effective natural language hypothesis proposal

Abstract

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.