Research map / Reinforcement learning / Visual and representation RL
Value-Consistent Representation Learning for Data-Efficient Reinforcement Learning
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 |
|---|---|---|---|
| Add value-consistency loss to SPR | SPR / Rainbow DQN · Atari-100K | IQM HNS (%): 33.7 -> 39.7 | Helped |
| Add value-consistency loss to SPR | SPR / Rainbow DQN · Atari-100K | Optimality Gap (%): 57.7 -> 54.4 | Helped |
| Add value-consistency loss to SPR | SPR / SAC · DeepMind Control 100K | IQM HNS: 700 -> 745 | Helped |
| Add value-consistency loss to SPR | SPR / SAC · DeepMind Control 100K | Optimality Gap: 335 -> 283 | Helped |
| Replace the VCR mixed-target distance with d_MSE (ablation) | VCR · Atari-100K | IQM HNS (%): 39.7% -> 30.9% | Hurt |
| Replace the VCR mixed-target distance with d_MSE-A (ablation) | VCR · Atari-100K | IQM HNS (%): 39.7% -> 34.0% | Hurt |
| Add reward prediction loss (ablation) | SPR · Atari-100K | IQM HNS (%): 33.7% -> 35.4% | Helped |
| Replace VCR loss on imagined state-action pairs with DQN loss (ablation) | SPR · Atari-100K | IQM HNS (%): 33.7% -> 27.2% | Hurt |
| Replace VCR loss with DQN loss and increase the DQN mini-batch (ablation) | SPR · Atari-100K | IQM HNS (%): 33.7% -> 17.6% | Hurt |
| Increase prediction steps from K=5 to K=9 (ablation) | VCR · Atari-100K | IQM HNS (%): 39.7% -> 39.7% | No clear effect |
| Set λ_VCR to 0, removing the value-consistency loss (ablation) | VCR · Atari-100K | IQM HNS: 39.7 -> 33.7 | Hurt |
| Increase λ_VCR from 0.2 to 1.0 (ablation) | VCR · Atari-100K | IQM HNS: 39.7 -> 28.8 | Hurt |
Closest papers by meaning
- The Value-Improvement Path: Towards Better Representations for Reinforcement Learning (AAAI 2021 · similarity 0.89)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Learning Robust Representations with Long-Term Information for Generalization in Visual Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Bridging State and History Representations: Understanding Self-Predictive RL (ICLR 2024 · similarity 0.88)
- VLMs-Guided Representation Distillation for Efficient Vision-Based Reinforcement Learning (CVPR 2025 · similarity 0.88)
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.