Research map / Reinforcement learning / Visual and representation RL
APS: Active Pretraining with Successor 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 |
|---|---|---|---|
| combine state-entropy exploration with successor-feature task conditioning | DrQ pretraining · Atari 26 | Mean Human-Norm'd: 70.4 -> 99.04 | Helped |
| combine state-entropy exploration with successor-feature task conditioning | DrQ pretraining · Atari 57 | Mean Human-Norm'd: 68.42 -> 103.04 | Helped |
| remove fine-tuning (ablation) | APS · Atari 26 | Mean Human-Normalized Score: 99.04 -> 81.41 | Hurt |
| remove fine-tuning (ablation) | APS · Atari 26 | Median Human-Normalized Score: 58.80 -> 49.18 | Hurt |
| add the APS fine-tuning scheme (ablation) | VISR · Atari 26 | Mean Human-Normalized Score | Helped |
| do not share the state encoder (ablation) | APS · Atari 26 | Mean Human-Normalized Score: 99.04 -> 87.59 | Hurt |
| do not share the state encoder (ablation) | APS · Atari 26 | Median Human-Normalized Score: 58.80 -> 51.45 | Hurt |
| train APT on data collected by APS rather than APT-collected data (ablation) | APT · Atari 26 dense-reward and exploration games | score | Helped |
| train VISR on data collected by APS rather than VISR-collected data (ablation) | VISR · Atari 26 | score | Helped |
Closest papers by meaning
- Towards Principled Unsupervised Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- SEMDICE: Off-policy State Entropy Maximization via Stationary Distribution Correction Estimation (ICLR 2025 · similarity 0.87)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.86)
- Beyond-Expert Performance with Limited Demonstrations: Efficient Imitation Learning with Double Exploration (ICLR 2025 · similarity 0.86)
- Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder (AAAI 2025 · similarity 0.86)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.86)
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.