AlphaZero

Google DeepMind

Structures found in this family (2)

Linear Subspace, Linear Direction

By model (2)

AlphaZero (chess, ResNet policy/value network, self-play)

Linear Direction

AlphaZero (Hex, ResNet policy/value network, self-play)

Linear Subspace

Observations (2)

AlphaZero-Hex cell embeddings grow to match board adjacency

AlphaZero's novel chess concepts are causally steerable and teachable

Papers

Evaluation Beyond Task Performance: Analyzing Concepts in AlphaZero in Hex (2022), Bridging the Human-AI Knowledge Gap: Concept Discovery and Transfer in AlphaZero (2023)