Research map / Reinforcement learning / Visual and representation RL
Self-Paced Context Evaluation for Contextual Reinforcement Learning
Theresa Eimer, André Biedenkapp, Frank Hutter, Marius Lindauer
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| Use value-function progress to choose curriculum instances instead of round-robin ordering | PPO on Ant · Ant | training steps to reach reward around 11: 10× as many steps | Improved |
| Use value-function progress to choose curriculum instances instead of round-robin ordering | DQN · CartPole | mean performance per episode: 25 -> around 150 | Improved |
| Use value-function progress to choose curriculum instances instead of round-robin ordering | SPaCE on BallCatching · BallCatching | training speed: a factor of at least 10 | Improved |
| Use value-function progress to choose curriculum instances instead of round-robin ordering | SPaCE on PointMass · PointMass, uniformly sampled instances | final performance | Improved |
| Use value-function progress to choose curriculum instances instead of round-robin ordering | SPaCE on mazes · 5x5 mazes | episodes to generalize across mazes: 10 episodes vs several hundred episodes | Improved |
| Target the entire instance space rather than focusing on generated hard instances | SPaCE vs SPDRL · PointMass, uniformly sampled test instances | learning speed and final performance | Improved |
| Replace value-progress selection with context-space distance (ablation) | SPaCE (context SPaCE) · PointMass | learning speed and overall performance | No clear effect |
| Use a larger curriculum increment κ across the tested values (ablation) | SPaCE on PointMass · PointMass test, after $10^6$ steps | mean reward | No clear effect |
| Vary the performance threshold η across the tested values (ablation) | SPaCE on PointMass · PointMass test, after $10^6$ steps | mean reward | No clear effect |
| Multiply the training-set size instead of adding instances incrementally (ablation) | SPaCE · PointMass | training speed | Hurt |
| Train on only 25 or 50 instances rather than larger sets (ablation) | SPaCE · PointMass test | test performance | Hurt |
| Use SPDRL's hard-instance distribution rather than a space-covering instance set (ablation) | SPDRL · PointMass, hard instances | performance over training: first ~200,000 steps | Mixed |
Closest papers by meaning
- In-Context Reinforcement Learning for Variable Action Spaces (ICML 2024 · similarity 0.87)
- XLand-100B: A Large-Scale Multi-Task Dataset for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Scaling Offline Evaluation of Reinforcement Learning Agents through Abstraction (AAAI 2024 · similarity 0.86)
- In-Context Decision Transformer: Reinforcement Learning via Hierarchical Chain-of-Thought (ICML 2024 · similarity 0.86)
- The Generalization Gap in Offline Reinforcement Learning (ICLR 2024 · similarity 0.86)
- Zero-Shot Context Generalization in Reinforcement Learning from Few Training Contexts (NeurIPS 2025 · 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.
