Research map / Reinforcement learning / Visual and representation RL
Understanding the Complexity Gains of Single-Task RL with a Curriculum
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 |
|---|---|---|---|
| add rollin-mixed initial distribution and previous-policy initialization (CURRICULUM, beta=0.75) | stochastic PG on four-room tabular navigation · Four-room navigation, hard and easy rewards | curriculum progress and final return | Helped |
| add goal relabeling variant combined with CURRICULUM (beta=0.1) (ablation) | goal reaching with relabeling on antmaze-umaze · antmaze-umaze, Δ=1/24 and 1/12 | learning progress κ: 0.89 vs 0.91; 0.66 vs 0.74 (uniform sampling) | No clear effect |
| add CURRICULUM (beta=0.1) to Go-Explore-style exploration noise, Δ=1/12, uniform sampling (ablation) | goal reaching on antmaze-umaze · antmaze-umaze Go-Explore Noise=0.1 | learning progress κ: 0.38 vs 0.38 | No clear effect |
| add CURRICULUM (beta=0.1) with Δ=1/12 in vanilla goal reaching, geometric sampling | antmaze-umaze goal reaching · antmaze-umaze vanilla Δ=1/12 | learning progress κ: 0.38 to 0.55 | Helped |
| add CURRICULUM (beta=0.1) to ant locomotion | non-goal locomotion ant · ant speed-tracking curriculum | learning progress κ at 0.5M steps: 0.72 to 0.82 | Helped |
| add CURRICULUM (beta=0.1) to hopper locomotion | non-goal locomotion hopper · hopper speed-tracking curriculum at 1.0M steps | return: 3386.2 to 3421.9 (within error) | No clear effect |
| add CURRICULUM (beta=0.1) to walker locomotion | non-goal locomotion walker · walker at 0.5M steps | learning progress κ: 0.83 to 0.79 at 0.5M; 0.92 to 0.94 at 1.0M | Mixed |
Closest papers by meaning
- Sample Efficient Myopic Exploration Through Multitask Reinforcement Learning with Diverse Tasks (ICLR 2024 · similarity 0.88)
- Progress Reward Model for Reinforcement Learning via Large Language Models (NeurIPS 2025 · similarity 0.88)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.87)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Diffusion-based Curriculum Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Learning Progress Driven Multi-Agent Curriculum (ICML 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.