AI Paper Atlas

Research map / Reinforcement learning / MDPs and planning

Anytime-Competitive Reinforcement Learning with Policy Prior

Jianyi Yang, Pengfei Li, Tongxin Li, Adam Wierman, Shaolei Ren

NeurIPS 2023 · Reinforcement learning › MDPs and planning · approach: Average-reward and regularized MDPs · 0 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
Apply ACD safe-action projection to the learned policyRL policy · Carbon-aware resource managementanytime-constraint violation rate: 27.5% -> zero violationImproved
Apply ACD safe-action projection to the learned policyRL policy · Carbon-aware resource managementanytime-constraint violation rate: 58.6% -> zero violationImproved
Train the policy in the ACD-induced MDP rather than the original environmentRL+ACD · Carbon-aware resource managementregretImproved
Use anytime cost constraints instead of expected cost constraintsSustainable AI inference · Sustainable AI inferenceregretHurt
Increase the constraint relaxation parameter λACRL · Sustainable AI inferenceregretImproved
Increase λ or b to relax the anytime constraintsACRL · Carbon-aware resource managementregretImproved
Use anytime-safe ACD action selection instead of unconstrained reward optimizationACRL · Carbon-aware resource managementanytime-constraint violation rateImproved
Use a randomly initialized ML policy as the ACD inputRandom+ACD · Carbon-aware resource managementregret: 61 to 64.51Hurt

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