Research map / Reinforcement learning / Visual and representation RL
MoCoDA: Model-based Counterfactual Data Augmentation
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 |
|---|---|---|---|
| Use a locally factored rather than an unfactored dynamics model | MoCoDA dynamics modeling · 2D Navigation, MoCoDA distribution | MSE × 1e2: 0.95 $\pm$ 0.06 -> 0.33 $\pm$ 0.11 | Helped |
| Use a locally factored rather than a globally factored dynamics model | MoCoDA dynamics modeling · 2D Navigation, MoCoDA distribution | MSE × 1e2: 0.41 $\pm$ 0.02 -> 0.33 $\pm$ 0.11 | Helped |
| Train on MoCoDA-augmented rather than empirical data | BCQ · 2D Navigation | Average steps to completion: 58.5 $\pm$ 10.1 -> 22.8 $\pm$ 0.4 | Helped |
| Train on MoCoDA-augmented rather than empirical data | CQL · 2D Navigation | Average steps to completion: 45.8 $\pm$ 4.0 -> 22.8 $\pm$ 0.2 | Helped |
| Train on MoCoDA-augmented rather than empirical data | TD3-BC · 2D Navigation | Average steps to completion: 40.0 $\pm$ 16.1 -> 21.0 $\pm$ 0.7 | Helped |
| Train on MoCoDA-augmented rather than empirical data | SAC · 2D Navigation | Average steps to completion: 53.1 $\pm$ 9.8 -> 38.8 $\pm$ 18.3 | Helped |
| Use MoCoDA-U rebalancing rather than MoCoDA sampling | CQL · 2D Navigation | Average steps to completion: 22.8 $\pm$ 0.2 -> 22.7 $\pm$ 0.3 | No clear effect |
| Train on MoCoDA rather than empirical data | BCQ · HookSweep2 | Average success rate: 2.0 $\pm$ 1.6 -> 20.7 $\pm$ 4.1 | Helped |
| Prioritize and rebalance block-position coverage with MoCoDA-P | BCQ · HookSweep2 | Average success rate: 20.7 $\pm$ 4.1 -> 64.7 $\pm$ 4.1 | Helped |
| Prioritize and rebalance block-position coverage with MoCoDA-P | TD3-BC · HookSweep2 | Average success rate: 38.7 $\pm$ 7.5 -> 84.0 $\pm$ 2.8 | Helped |
| Use a locally factored rather than a fully connected dynamics model (ablation) | MoCoDA with BCQ · HookSweep2 | Average success rate: 14.0 $\pm$ 3.3 -> 20.7 $\pm$ 4.1 | Helped |
| Use a locally factored rather than a fully connected dynamics model (ablation) | MoCoDA-P with BCQ · HookSweep2 | Average success rate: 15.3 $\pm$ 4.1 -> 64.7 $\pm$ 4.1 | Helped |
Closest papers by meaning
- Causal Action Influence Aware Counterfactual Data Augmentation (ICML 2024 · similarity 0.88)
- Modeling Unseen Environments with Language-guided Composable Causal Components in Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Localized Dynamics-Aware Domain Adaption for Off-Dynamics Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.87)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.87)
- Temporal Distance-aware Transition Augmentation for Offline Model-based Reinforcement Learning (ICML 2025 · similarity 0.87)
- Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction (ICLR 2025 · 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.