Research map / Reinforcement learning / Visual and representation RL
A Probabilistic Interpretation of Self-Paced Learning with Applications to Reinforcement Learning
Pascal Klink, Hany Abdulsamad, Boris Belousov, Carlo D'Eramo, Jan Peters, Joni Pajarinen
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 V_LB performance-threshold alpha schedule instead of K_alpha/zeta (ablation) | SPDL · Point-mass 3D (PPO) | iterations to reach threshold: 165 ± 14 -> 120 ± 10 | Improved |
| use V_LB alpha schedule instead of K_alpha/zeta (ablation) | SPDL · Point-mass 3D (SAC) | iterations to reach threshold: 94 ± 8 -> 67 ± 4 | Improved |
| use V_LB alpha schedule instead of K_alpha/zeta (ablation) | SPDL · Point-mass 2D (SAC) | iterations to reach threshold: 134 ± 30 -> 59 ± 3 | Improved |
| use V_LB alpha schedule instead of K_alpha/zeta (ablation) | SPDL · Ball-catching (SAC) | iterations to reach threshold: 205 ± 12 -> 139 ± 8 | Improved |
| use V_LB alpha schedule instead of K_alpha/zeta (ablation) | SPDL · Ant (PPO) | final reward: 1371 ± 23 -> 1305 ± 38 | No clear effect |
| use V_LB alpha schedule instead of K_alpha/zeta (ablation) | SPDL · Point-mass 2D (SAC) | final reward: 5.67 ± 0.77 -> 6.56 ± 0.82 | No clear effect |
| increase context dimension from 2D to 3D (add friction) | SPDL · Point-mass (PPO) | final reward: 8.94 ± 0.10 -> 8.45 ± 0.4 | No clear effect |
| increase context dimension from 2D to 3D (add friction) | GoalGAN · Point-mass (PPO) | final reward: 1.63 ± 0.5 -> 0.66 ± 0.1 | Hurt |
| increase context dimension from 2D to 3D (add friction) | ALP-GMM · Point-mass (PPO) | final reward: 5.23 ± 0.4 -> 2.43 ± 0.3 | Hurt |
| initialize context distribution to tasks the initial policy solves | SPDL · Ball-catching (PPO) | final reward: 53.9 ± 0.4 -> 49.3 ± 1.4 | Hurt |
| initialize context distribution to tasks the initial policy solves | GoalGAN · Ball-catching (PPO) | final reward: 42.6 ± 2.7 -> 45.9 ± 1.0 | Improved |
| use SPDL curriculum instead of default learning without curriculum | PPO · Ball-catching | final reward: 22.1 ± 0.3 -> 53.9 ± 0.4 | Improved |
Closest papers by meaning
- Self-Paced Deep Reinforcement Learning (NeurIPS 2020 · similarity 0.93)
- Curriculum Reinforcement Learning via Constrained Optimal Transport (ICML 2022 · similarity 0.90)
- Self-Paced Context Evaluation for Contextual Reinforcement Learning (ICML 2021 · similarity 0.90)
- Automated curriculum generation through setter-solver interactions (ICLR 2020 · similarity 0.90)
- Curriculum Design for Trajectory-Constrained Agent: Compressing Chain-of-Thought Tokens in LLMs (NeurIPS 2025 · similarity 0.88)
- Diffusion-based Curriculum Reinforcement Learning (NeurIPS 2024 · 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.
