Research map / Reinforcement learning / Visual and representation RL
Mask-based Latent Reconstruction for 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 MLR masked latent reconstruction objective | pixel-based RL baseline · DMControl-100k | mean score: 616.8 -> 772.8 | Helped |
| add MLR objective | Rainbow baseline · Atari-100k | IQM: 0.292 -> 0.432 | Helped |
| add MLR objective at 500k steps | pixel-based RL baseline · DMControl-500k | mean score: 856.3 -> 896.5 | Helped |
| use spatial-temporal cube masking instead of spatial-only (ablation) | MLR · DMControl-100k | mean score: 727.3 -> 767.3 | Helped |
| use spatial-temporal cube masking instead of temporal-only (ablation) | MLR · DMControl-100k | mean score: 688.5 -> 767.3 | Helped |
| reconstruct in latent space instead of pixel space (ablation) | MLR · DMControl-100k | mean score: 647.2 -> 767.3 | Helped |
| add action tokens to the latent decoder input (ablation) | MLR · DMControl-100k | mean score: 694.2 -> 767.3 | Helped |
| mask pixels instead of conv features (ablation) | MLR · DMControl-100k | mean score: 727.2 -> 767.3 | Helped |
| add momentum latent decoder for target states (ablation) | MLR · DMControl-100k | mean score: 767.3 -> 706.4 | Hurt |
| increase predictive decoder depth from 2 to 8 layers (ablation) | MLR · DMControl-100k | mean score: 767.3 -> 728.3 | Hurt |
| use cosine similarity loss instead of MSE (ablation) | MLR · DMControl-100k | mean score | Helped |
Closest papers by meaning
- Masked Autoencoding for Scalable and Generalizable Decision Making (NeurIPS 2022 · similarity 0.89)
- Stochastic Latent Actor-Critic: Deep Reinforcement Learning with a Latent Variable Model (NeurIPS 2020 · similarity 0.89)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.88)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.86)
- Masking Teacher and Reinforcing Student for Distilling Vision-Language Models (CVPR 2026 · similarity 0.86)
- MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation (NeurIPS 2026 · similarity 0.86)
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.