Research map / Reinforcement learning / Visual and representation RL
Grounding Aleatoric Uncertainty for Unsupervised Environment Design
Minqi Jiang, Michael D Dennis, Jack Parker-Holder, Andrei Lupu, Heinrich Kuttler, Edward Grefenstette, Tim Rocktäschel, Jakob Nicolaus Foerster
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| replace PLR-perp with SAMPLR (fictitious transitions) | PLR-perp · CarRacingF1, q=0.4 | episodic return: 43 -> 195 | Improved |
| replace PLR-perp with SAMPLR | PLR-perp · CarRacingF1, q=0.6 | episodic return: -115 -> -1 | Improved |
| replace PLR-perp with SAMPLR | PLR-perp · CarRacingF1, q=0.8 | episodic return: -151 -> -111 | Improved |
| replace PLR-perp with SAMPLR | PLR-perp · CarRacingF1, q=0.2 | episodic return: 323 -> 393 | Improved |
| replace naive grounding with SAMPLR | PLR-perp with naive grounding · CarRacingF1, q=0.4 | episodic return: 75 -> 195 | Improved |
| replace naive grounding with SAMPLR | PLR-perp with naive grounding · CarRacingF1, q=0.6 | episodic return: -79 -> -1 | Improved |
| replace naive grounding with SAMPLR | PLR-perp with naive grounding · CarRacingF1 ground truth q~Beta(1,15) | episodic return: 618 -> 616 | No clear effect |
| replace DR with SAMPLR | DR · CarRacingF1, q=0.4 | episodic return: 94.7 -> 195 | Improved |
| replace DR with SAMPLR | DR · CarRacingF1 ground truth | episodic return: 581 -> 616 | Improved |
| use plain PLR-perp curriculum instead of DR (no grounding) | DR · CarRacingF1 ground truth | episodic return: 581 -> 543 | Hurt |
| add naive grounding to PLR-perp | PLR-perp · CarRacingF1 ground truth | episodic return: 543 -> 618 | Improved |
| add naive grounding to PLR-perp | PLR-perp · Stochastic Fruit Choice | return | Improved |
Closest papers by meaning
- No Regrets: Investigating and Improving Regret Approximations for Curriculum Discovery (NeurIPS 2024 · similarity 0.91)
- Improving Regret Approximation for Unsupervised Dynamic Environment Generation (NeurIPS 2025 · similarity 0.90)
- Adversarial Environment Design via Regret-Guided Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Improving Environment Novelty Quantification for Effective Unsupervised Environment Design (NeurIPS 2024 · similarity 0.88)
- Marginal Benefit Driven RL Teacher for Unsupervised Environment Design (AAAI 2025 · similarity 0.88)
- Refining Minimax Regret for Unsupervised Environment Design (ICML 2024 · 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.
