An independent study

Foundations of
Efficient Intelligence

Why is biological intelligence so computationally and sample efficient — and how do we build AI systems with those properties? An interactive graduate textbook built one paper, one lecture at a time, drawing on computational cognitive science, neuroscience, ML, and systems.

Samuel Gershman Lab

Strategy reuse, reinforcement learning, memory, Bayesian cognition, human decision-making.

01Neural evidence that humans reuse strategies to solve new tasks
02Policy Compression Information Bottleneck
ArticleSlidesPaper
03Action subsampling supports policy compression in large action spaces
04Action Chunking Policy Compression
ArticleSlidesPaper
05Key Value Memory In The Brain
ArticleSlidesPaper
06Synthesizing World Models Bilevel Planning
ArticleSlidesPaper

Josh Tenenbaum Lab

One-shot learning, program induction, concept learning, causal reasoning, intuitive physics.

01Human Level Concept Learning
ArticleSlidesPaper
02Bayesian Models Conceptual Development
ArticleSlidesPaper
03Rapid Trial And Error Tool Use
ArticleSlidesPaper
04Bayesian Program Induction Language
ArticleSlidesPaper
05Building Machines Learn Think Like People
ArticleSlidesPaper

Karl Friston

Predictive coding, active inference, the free energy principle.

01Theory Of Cortical Responses
ArticleSlidesPaper
02Free Energy Principle Active Inference
ArticleSlidesPaper

Danijar Hafner

Latent world models, Dreamer, planning.

01Mastering Diverse Control World Models
ArticleSlidesPaper
02Dino Wm
ArticleSlidesPaper

Eric Schulz Lab

Meta-learned models of cognition; neural networks and LLMs as models of human learning.

01Meta Learned Models Of Cognition
ArticleSlidesPaper

Matthew Botvinick Lab

Deep reinforcement learning as a theory of brain and cognition; meta-RL, distributional value coding.

01Learning To Reinforcement Learn
ArticleSlidesPaper