Research map / Reinforcement learning / Imitation learning
Provable Representation Learning for Imitation with Contrastive Fourier Features
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 |
|---|---|---|---|
| Learn representations with contrastive Fourier features before behavioral cloning | Behavioral cloning with an Atari CNN · Atari 2600 (60 games) | normalized improvement over vanilla BC: >40% on over half of the games | Helped |
| Learn representations with an energy-based contrastive model | Behavioral cloning with an Atari CNN · Atari 2600 (60 games) | normalized improvement over vanilla BC: >40% on over half of the games | Helped |
| Use contrastive Fourier features instead of vanilla behavioral cloning | Tabular behavioral cloning · Tree environments with low-rank transitions | average performance | Helped |
| Use SVD features instead of contrastive Fourier features (ablation) | Representation learning for tabular behavioral cloning · Tree environments with low-rank transitions | average performance | Mixed |
| Increase the state embedding size from 256 to 512 (ablation) | Contrastive representation learning · Atari 2600 (60 games) | normalized improvement over vanilla BC | Helped |
| Use more offline transitions, reducing the difference between offline data and target demonstrations (ablation) | Contrastive representation learning · Atari 2600 (60 games) | normalized improvement over vanilla BC | Helped |
| Use a latent-space model objective (DeepMDP or DBC) instead of contrastive learning (ablation) | Behavioral cloning with learned representations · Atari 2600 (60 games) | normalized improvement over vanilla BC | Hurt |
| Allow behavioral-cloning gradients to update the latent-space model representation (ablation) | DeepMDP and DBC · Atari 2600 (60 games) | normalized improvement over vanilla BC | Helped |
| Ignore the Fourier-feature scaling factor E(s,a) (ablation) | Contrastive Fourier features · Experiments | performance | No clear effect |
| Use contrastive Fourier features rather than SVD representations (ablation) | Tabular representation learning · Tree environments with reduced offline data or larger state spaces | average performance | Helped |
Closest papers by meaning
- Robust Visual Imitation Learning with Inverse Dynamics Representations (AAAI 2024 · similarity 0.89)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Imitation from Observations with Trajectory-Level Generative Embeddings (NeurIPS 2026 · similarity 0.88)
- OLLIE: Imitation Learning from Offline Pretraining to Online Finetuning (ICML 2024 · similarity 0.88)
- Agnostic Interactive Imitation Learning: New Theory and Practical Algorithms (ICML 2024 · similarity 0.87)
- Diffusion Imitation from Observation (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.