AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Inverse Preference Learning: Preference-based RL without a Reward Function

Joey Hejna, Dorsa Sadigh

NeurIPS 2023 · Reinforcement learning › RL from feedback · approach: Theory of RLHF · 6 citations (OpenAlex)

Read the paper

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
remove regularization (lambda = 0) (ablation)IPL · hop-m-rnormalized score: 73.57 -> 10.41Hurt
remove regularization (lambda = 0) (ablation)IPL · walk-m-rnormalized score: 59.92 -> 4.85Hurt
remove regularization (lambda = 0) (ablation)IPL · can-mhnormalized score: 57.6 -> 13.8Hurt
remove data augmentation (ablation)IPL · hop-m-rnormalized score: 73.57 -> 70.46Hurt
remove data augmentation (ablation)IPL · hop-m-enormalized score: 74.52 -> 51.26Hurt
remove data augmentation (ablation)IPL · walk-m-enormalized score: 108.51 -> 108.91No clear effect
remove data augmentation (ablation)IPL · can-phnormalized score: 74.8 -> 68.6Hurt
remove data augmentation (ablation)IPL · lift-phnormalized score: 97.60 -> 98.0No clear effect
replace IQL with XQL (ablation)IPL · hop-m-rnormalized score: 73.57 -> 80.4Improved
replace IQL with XQL (ablation)IPL · walk-m-enormalized score: 108.51 -> 75.16Hurt
replace IQL with XQL (ablation)IPL · hop-m-enormalized score: 74.52 -> 54.3Hurt
use IPL instead of explicit Markovian reward model (MR)IQL · MetaWorld Button Press, 4000 queriessuccess rate: 78.3 -> 90.2Improved

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