Research map / Label-efficient and robust learning / Recognition and long-tail
Planning from Pixels in Atari with Learned Symbolic Representations
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 |
|---|---|---|---|
| Replace B-PROST features with learned VAE binary features in RolloutIW | RolloutIW(1) · Atari 55 games | games with best score: 14 -> 39 | Helped |
| Replace B-PROST features with learned VAE binary features in RolloutIW | RolloutIW(1) · Atari 55 games | games scoring above human: 22 -> 25 | Helped |
| Use a compact learned feature set instead of B-PROST | RolloutIW(1) · Atari | feature count: 20,598,848 -> 4,500 | Helped |
| Change latent-space shape to 4 × 4 × 200 (ablation) | VAE-IW · Atari 47 games | games with best score: 29 -> 20 | Mixed |
| Replace discrete latents with continuous latents quantized to 4 bits (ablation) | VAE-IW · Atari 47 games | games with best score: 23 -> 18 | Mixed |
| Replace discrete latents with continuous latents quantized to 6 bits (ablation) | VAE-IW · Atari 47 games | games with best score: 23 -> 10 | Mixed |
| Increase β from 10⁻⁴ to 10⁻³ (ablation) | VAE-IW · Atari 47 games | games with best score: 29 -> 19 | Mixed |
| Add risk aversion by amplifying negative rewards (ablation) | VAE-IW · Atari 47 games | games with best score: 16 -> 32 | Helped |
| Increase planning budget from 0.5s to 32s (ablation) | VAE-IW · Atari 47 games | games with best score: 11 -> 37 | Helped |
| Increase planning budget from 0.5s to 32s (ablation) | VAE-IW · Atari 47 games | games where VAE-IW has higher average score than B-PROST RolloutIW: 73% -> 62% | Mixed |
Closest papers by meaning
- Planning With Pixels in (Almost) Real Time (AAAI 2018 · similarity 0.88)
- Width-based Lookaheads with Learnt Base Policies and Heuristics Over the Atari-2600 Benchmark (NeurIPS 2021 · similarity 0.85)
- Learning to Play Atari in a World of Tokens (ICML 2024 · similarity 0.84)
- Unsupervised State Representation Learning in Atari (NeurIPS 2019 · similarity 0.84)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.83)
- Learning Successor Features the Simple Way (NeurIPS 2024 · similarity 0.81)
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.