Survey Examines Programs as a Universal Language for Concepts

Summary

Humans can learn and generalize new concepts from sparse data because they represent knowledge in rich structural forms. This survey examines the proposal that programs could serve as a universal representation of concepts. It reviews computational concept-learning models that use programs as their concept representation and assesses how much they contribute toward a universal representational language. The paper was originally completed as an M.S. capstone project at UCLA in 2022 and is categorized in artificial intelligence, programming languages, and symbolic computation.