AI Paper Atlas

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

ULTHO: Ultra-Lightweight yet Efficient Hyperparameter Optimization in Deep Reinforcement Learning

Mingqi Yuan, Bo Li, Xin Jin, Wenjun Zeng

ICCV 2025 · Reinforcement learning › Policy gradient and actor-critic · approach: Regularized scalable deep value learning · 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
add ULTHO online HP selectionPPO · Procgen (full level distribution)aggregated mean: 40.42 -> 56.0Improved
add ULTHO online HP selectionPPO · Procgen (full level distribution)aggregated IQM: 38.41 -> 56.41Improved
add relay optimization on top of ULTHO (ablation)ULTHO · Procgen (full level distribution)aggregated IQM: 56.41 -> 49.01Hurt
add relay optimization on top of ULTHO (ablation)ULTHO · ALE-5 and stationary environmentsnormalized scoreImproved
use ULTHO instead of PPO baselinePPO · MiniGrid (sparse rewards)returnNo clear effect
use ULTHOPPO · PyBulletreturnImproved
vary exploration coefficient c and window W (ablation)ULTHO · Procgenaggregated returnNo clear effect
add more HP clusters (ablation)ULTHO · ProcgenreturnImproved
use ULTHO with equal 100M step budgetPPO+ULTHO · ALE-5normalized scoreImproved
apply ULTHO to off-policy SACSAC · DMCreturnMixed
use ULTHO vs TMIHFPPO · Procgen (full level distribution)aggregated IQM: 48.7 -> 56.41Improved

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