AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Learning from Active Human Involvement through Proxy Value Propagation

Zhenghao Peng, Wenjie Mo, Chenda Duan, Quanyi Li, Bolei Zhou

NeurIPS 2023 · Reinforcement learning › Policy gradient and actor-critic · approach: Model-based RL with guarantees · 1 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
disable TD learning (J_TD(Q)=0) (ablation)PVP · MetaDrive-Keyboardsuccess rate: 0.857 -> 0.220Hurt
add environmental reward (ablation)PVP · MetaDrive-Keyboardsuccess rate: 0.857 -> 0.755Hurt
add environmental reward (ablation)PVP · CARLAroute completion: 0.92 -> 0.793Hurt
remove balanced buffer (ablation)PVP · CARLAroute completion: 0.92 -> 0.62Hurt
remove novice buffer (ablation)PVP · CARLAroute completion: 0.92 -> 0.708Hurt
use stochastic policy for novice (ablation)PVP · CARLAroute completion: 0.92 -> 0.40Hurt
use stochastic policy for novice (ablation)PVP · CARLAhuman data usage: 6.6K -> 12.3KHurt
replace PVP objective with vanilla CQL objective (ablation)PVP · CARLAsuccess rate: 0.73 -> 0.266Hurt
make HACO policy deterministic (ablation)HACO · CARLAsuccess rate: 0.40 -> 0.20No clear effect
replace HACO with PVPPVP · MetaDrive-Keyboardepisodic return: 143.287 -> 353.636Improved
replace HACO with PVPPVP · CARLAroute completion: 0.52 -> 0.92Improved
replace HACO with PVP in human data usagePVP · MetaDrive-Keyboardhuman data usage: 19.2K -> 14.6KImproved

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