AI Paper Atlas

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

Factored Policy Gradients: Leveraging Structure for Efficient Learning in MOMDPs

Thomas Spooner, Nelson Vadori, Sumitra Ganesh

NeurIPS 2021 · Reinforcement learning › Policy gradient and actor-critic · approach: Trajectory reuse in policy gradients · 2 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
use factor baselines that exclude targets unrelated to each policy factorpolicy gradients on the search bandit · Search bandit, λ = 0learning stability and convergenceImproved
add an action-dependent baseline b(s, a) (ablation)VPG · Search bandit, λ = 0wall-clock training speed: 10534 -> 80 iterations per secondHurt
add a state-dependent baseline b(s) (ablation)FPG · Search bandit, λ = 0wall-clock training speed: 9950 -> 9670 iterations per secondHurt
use factor baselines with coupled targetsFPG · Search bandit, λ > 0variance reductionImproved
replace the global reward with a local spatial baseline that excludes unconnected street reward termsshared-policy PPO with GAE · 3×3 traffic gridlearning efficiencyImproved
expand the local baseline to include lights farther away (ablation)shared-policy PPO with GAE · 2×6 traffic gridterminal performanceHurt
use a joint policy instead of a shared policy (ablation)PPO with GAE · 3×3 traffic gridperformanceMixed
remove factor baselines and use vanilla policy gradients (ablation)policy gradients · Search banditvarianceHurt

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