Research map / Reinforcement learning / Imitation learning
Data augmentation for efficient learning from parametric experts
Alexandre Galashov, Josh Merel, Nicolas Heess
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 |
|---|---|---|---|
| Add expert re-querying of actions at perturbed states (APC) instead of keeping the original action (Naive ABC) (ablation) | Behavioral cloning · Offline cloning on Humanoid Run, Humanoid Walls, Insert Peg | episodic reward vs number of trajectories | Improved |
| Add state perturbation with the same action (Naive ABC) to plain BC (ablation) | Behavioral cloning · Offline cloning on all tasks | episodic reward | No clear effect |
| Shrink the student network torso below [256, 256, 256] (expert compression) (ablation) | APC vs BC · Expert compression | episodic reward | Mixed |
| Add APC to cloning from a privileged expert with vision-based student (ablation) | Behavioral cloning · Insert Peg with privileged target position vs vision | episodic reward | Improved |
| Use short trajectories (200 steps) plus one full trajectory instead of full trajectories (ablation) | BC and Naive ABC · Humanoid Run and Humanoid Walls | episodic reward | Hurt |
| Use short trajectories (200 steps) instead of full trajectories (ablation) | BC and Naive ABC · Insert Peg | episodic reward | Improved |
| Use short trajectories instead of full ones (ablation) | APC · All three tasks | episodic reward | No clear effect |
| Add APC state-based augmentation to DAgger (beta 0 and 0.3) (ablation) | DAgger · Humanoid Run and Walls | episodic reward | Improved |
| Add image random-crop augmentation to DAgger (ablation) | DAgger on Humanoid Walls · Humanoid Walls | episodic reward | Improved |
| Use analytical cross-entropy instead of log-probability of expert mean action (ablation) | DAgger · Humanoid Run and Walls | episodic reward | Improved |
| Add APC to kickstarting (ablation) | Kickstarting · Humanoid Run | episodic reward | Improved |
| Add APC to kickstarting (ablation) | Kickstarting · Humanoid Walls | episodic reward | No clear effect |
Closest papers by meaning
- Split the Differences, Pool the Rest: Provably Efficient Multi-Objective Imitation (NeurIPS 2026 · similarity 0.88)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.88)
- Behavior Cloning is Not All You Need: The Optimality of On-Policy Distillation for Noisy Expert Feedback (NeurIPS 2026 · similarity 0.88)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.88)
- When Does Interaction Help? Representational Tradeoffs in Value-Based Imitation Learning (NeurIPS 2026 · similarity 0.87)
- Interactive and Hybrid Imitation Learning: Provably Beating Behavior Cloning (NeurIPS 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.
