LEARNING TO LEARN CONCEPTS AS PROGRAMS: HIERARCHICAL BAYES AND AMORTISED INFERENCE
Abstract
The research will seek to build systems that can automatically develop human-like expertise in a novel domain, which can reason flexibly about a new family of problems and communicate their reasoning in human-comprehensible language. The researchers believe that building such systems requires learning highly structured models of the world—models which exploit such principles as modularity, abstraction, hierarchy, composition, and probabilistic reasoning. These principles are currently best embodied in "programming languages"; therefore, the challenge to build more flexible and explainable AI systems should be framed as Program Learning ("programming to learn to learn").
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 20, 2023
- Source ID
- FA95502210387
Entities
People
- Joshua B. Tenenbaum
Organizations
- Air Force Office of Scientific Research
- Massachusetts Institute of Technology
- United States Air Force