Research map / Reinforcement learning / Policy gradient and actor-critic
Learning Self-Imitating Diverse Policies
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 self-imitation rewards from high-return replay trajectories to PPO | PPO · Humanoid episodic rewards | return: 532 -> 3602 | Helped |
| add self-imitation rewards from high-return replay trajectories to PPO | PPO · Walker episodic rewards | return: 252 -> 2996 | Helped |
| add self-imitation rewards from high-return replay trajectories to PPO | PPO · Walker dense rewards | return: 3401 -> 3263 | Hurt |
| add self-imitation rewards from high-return replay trajectories to PPO | PPO · Reacher episodic rewards | return: -12 -> -12 | No clear effect |
| use diverse-policy SVPG exploration with a Jensen-Shannon kernel instead of independent self-imitation agents | Self-Imitation ensemble · Sparse locomotion tasks | return | Helped |
| replace the Jensen-Shannon kernel with an RBF kernel (ablation) | SI-interact-JS · Sparse locomotion tasks | return | Hurt |
| reduce the self-imitation weight from ν=0.8 to ν=0.2 (ablation) | Self-Imitation · Humanoid episodic rewards | return: 3602 -> 395 | Hurt |
| increase the self-imitation weight from ν=0.8 to ν=1 (ablation) | Self-Imitation · Humanoid episodic rewards | return: 3602 -> 3891 | Helped |
| reduce the self-imitation weight from ν=0.8 to ν=0.2 (ablation) | Self-Imitation · Hopper episodic rewards | return: 2618 -> 481 | Hurt |
| increase the self-imitation weight from ν=0.8 to ν=1 (ablation) | Self-Imitation · Hopper episodic rewards | return: 2618 -> 2633 | Helped |
| increase replay-memory capacity from C=10 to C=50 (ablation) | Self-Imitation · Humanoid episodic rewards | return: 3602 -> 4159 | Helped |
| increase replay-memory capacity from C=10 to C=25 (ablation) | Self-Imitation · Hopper episodic rewards | return: 2618 -> 1624 | Hurt |
Closest papers by meaning
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.89)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.89)
- Learning Multimodal Behaviors from Scratch with Diffusion Policy Gradient (NeurIPS 2024 · similarity 0.89)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.88)
- Dual RL: Unification and New Methods for Reinforcement and Imitation Learning (ICLR 2024 · similarity 0.88)
- Beyond-Expert Performance with Limited Demonstrations: Efficient Imitation Learning with Double Exploration (ICLR 2025 · 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.