Research map / Reinforcement learning / Visual and representation RL
Multi-Task Deep Reinforcement Learning with PopArt
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Apply task-specific PopArt adaptive normalization to IMPALA updates | IMPALA · Atari-57, reward clipping | median human normalised score: 60% -> 110% | Helped |
| Apply task-specific PopArt adaptive normalization to IMPALA updates | IMPALA · Atari-57, unclipped rewards | median human normalised score: 101% versus close to 0% | Helped |
| Apply task-specific PopArt adaptive normalization to IMPALA updates | IMPALA · DmLab-30 | mean capped human normalised score: 49.4% -> 72.8% | Helped |
| Use task-specific value predictions without adaptive normalization (ablation) | IMPALA · Atari-57, with and without reward clipping | median human normalised score | Hurt |
| Add pixel control auxiliary learning (ablation) | PopArt-IMPALA · DmLab-30 | mean capped human normalised score: ∼2B frames | Helped |
| Add pixel control auxiliary learning (ablation) | IMPALA · DmLab-30 | mean capped human normalised score | Helped |
| Use the expanded action set with finer horizontal and vertical rotations (ablation) | IMPALA · DmLab-30 | mean capped human normalised score | Helped |
| Tune hyperparameters with a PBT population of 24 rather than 8 (ablation) | PopArt-IMPALA · DmLab-30 | mean capped human normalised score | No clear effect |
| Tune hyperparameters with a PBT population of 24 rather than 8 (ablation) | IMPALA · DmLab-30 | mean capped human normalised score | No clear effect |
Closest papers by meaning
- Investigating Multi-task Pretraining and Generalization in Reinforcement Learning (ICLR 2023 · similarity 0.91)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.89)
- Sharing Knowledge in Multi-Task Deep Reinforcement Learning (ICLR 2020 · similarity 0.88)
- Reinforcement Learning with Unsupervised Auxiliary Tasks (ICLR 2017 · similarity 0.88)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.87)
- Hard Tasks First: Multi-Task Reinforcement Learning Through Task Scheduling (ICML 2024 · similarity 0.87)
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.