Research map / Reinforcement learning / Visual and representation RL
Discovery of Options via Meta-Learned Subgoals
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 switching cost c=0.05 during option discovery (ablation) | MODAC · Gridworld transfer | average transfer performance: 0.05 | Helped |
| use MODAC-discovered options instead of primitive actions only (Flat) | manager on test tasks · Gridworld | learning speed on test tasks | Helped |
| MODAC options instead of MLSH or Option-Critic options | manager on test tasks · Gridworld | learning speed on test tasks | Helped |
| MODAC options instead of Flat agent | manager on test tasks · DeepMind Lab test task sets | asymptotic performance | Mixed |
| MODAC options instead of MLSH and Option-Critic options | manager on test tasks · DeepMind Lab | transfer performance | Helped |
| discover options from only 5M or 10M samples instead of 200M (ablation) | MODAC · DeepMind Lab one task set | transfer learning speed incl. training samples | Helped |
| transfer MODAC options discovered from pixel-control tasks to Atari | manager · Atari 4 games | learning speed on game score | Helped |
| MODAC options instead of Flat on MsPacman | manager · Atari MsPacman | learning speed and asymptote | Mixed |
| ignore episode terminations in pixel-control training tasks (ablation) | MODAC option discovery · Atari 4 games | learning curve of transfer | Mixed |
| number of options K=4 vs K=2 or K=8 (ablation) | MODAC · Gridworld | option quality (qualitative) | Mixed |
Closest papers by meaning
- Reward-Respecting Subtasks for Model-Based Reinforcement Learning (Abstract Reprint) (AAAI 2024 · similarity 0.90)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.89)
- Discovering Temporally-Aware Reinforcement Learning Algorithms (ICLR 2024 · similarity 0.89)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.88)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.88)
- Can Learned Optimization Make Reinforcement Learning Less Difficult? (NeurIPS 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.