AI Paper Atlas

Research map / Large language models / Agents, code and math

Using natural language and program abstractions to instill human inductive biases in machines

Sreejan Kumar, Carlos G Correa, Ishita Dasgupta, Raja Marjieh, Michael Hu, Robert D. Hawkins, Jonathan Cohen, Nathaniel Daw, Karthik R Narasimhan, Thomas L. Griffiths

NeurIPS 2022 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Ground the agent with human-generated language descriptionsPPO meta-RL agent · GSP boardswhite tiles revealed: p<0.0001Improved
Ground the agent with human-generated language descriptionsPPO meta-RL agent · Control boardswhite tiles revealed: p=0.021Hurt
Ground the agent with a flattened-board autoencoder targetPPO meta-RL agent · GSP boardswhite tiles revealed: p<0.001Improved
Ground the agent with a flattened-board autoencoder targetPPO meta-RL agent · GSP boards vs control boardsperformance difference: p=0.252No clear effect
Ground the agent with synthetic descriptionsPPO meta-RL agent · GSP boards and control boardswhite tiles revealed: p=0.0016Mixed
Use library-learning program embeddings instead of program embeddings without library learning (ablation)PPO meta-RL agent · GSP boardswhite tiles revealed: p<0.0001Improved
Ground the agent with program embeddings without library learning (ablation)PPO meta-RL agent · GSP boards vs control boardsperformance difference: p=0.11No clear effect
Add library-learned abstractions to program representations (ablation)DreamCoder program representations · GSP boardsprogram description lengthImproved
Use human-generated rather than synthetic language descriptions (ablation)Language representations · GSP boardsdescription lengthImproved
Use library-learning program description lengths rather than program description lengths without library learning (ablation)DreamCoder program representations · GSP boardscorrelation with human-language description length: p=0.0034Improved
Ground the agent with a DreamCoder recognition-model representation rather than a generic DNN representation (ablation)PPO meta-RL agent · GSP boards vs control boardsselective performance favoring GSP boardsImproved
Train a generic meta-RL agent without groundingPPO meta-RL agent · GSP boards vs control boardswhite tiles revealed: p<0.0001Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy