Research map / Reinforcement learning / Visual and representation RL
Masked Autoencoding for Scalable and Generalizable Decision Making
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 |
|---|---|---|---|
| closed-loop instead of open-loop execution | MaskDP · Walker walk single-goal, single-task pretraining | L2 distance: 5.822 -> 2.63 | Helped |
| closed-loop instead of open-loop execution | MaskDP · Quadruped run single-goal, single-task pretraining | L2 distance: 13.753 -> 12.912 | Helped |
| closed-loop instead of open-loop execution | MaskDP · Cheetah run single-goal, multi-task pretraining | L2 distance: 1.924 -> 1.939 | No clear effect |
| MaskDP instead of Goal-GPT | MaskDP closed-loop · Quadruped run single-goal, single-task | L2 distance: 18.313 -> 12.912 | Helped |
| MaskDP instead of Goal-MLP | MaskDP closed-loop · Cheetah run single-goal, single-task | L2 distance: 1.377 -> 1.385 | Mixed |
| use a mixed mask ratio instead of a fixed ratio (ablation) | MaskDP · goal reaching | L2 distance | Helped |
| fixed mask ratio of 95% (ablation) | MaskDP · goal reaching | L2 distance | Hurt |
| fixed mask ratio of 15% (BERT-style) (ablation) | MaskDP · goal reaching | L2 distance | Hurt |
| fixed mask ratio of 50% (ablation) | MaskDP · goal reaching | L2 distance | Mixed |
| train with total loss instead of masked loss only (ablation) | MaskDP · expert dataset pretraining | convergence speed | Helped |
| provide multiple future goals (foresight) in multi-goal reaching (ablation) | MaskDP closed-loop · multi-goal reaching | L2 distance | Helped |
| enlarge model from 3 to 5 attention layers (ablation) | MaskDP · Quadruped goal reaching | L2 distance | Helped |
Closest papers by meaning
- Self-Guided Masked Autoencoders for Domain-Agnostic Self-Supervised Learning (ICLR 2024 · similarity 0.87)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.87)
- Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking (NeurIPS 2024 · similarity 0.87)
- M^3PC: Test-time Model Predictive Control using Pretrained Masked Trajectory Model (ICLR 2025 · similarity 0.87)
- Understanding and Enhancing Mask-Based Pretraining towards Universal Representations (NeurIPS 2025 · similarity 0.86)
- Learning Versatile Skills with Curriculum Masking (NeurIPS 2024 · 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.