Research map / Reinforcement learning / Visual and representation RL
Dynamic-Depth Context Tree Weighting
Joao V Messias, Shimon Whiteson
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 |
|---|---|---|---|
| dynamic-depth fringe expansion (D2-CTW) instead of fixed-depth CTW | context tree weighting on Calgary Corpus · Calgary Corpus, unbounded memory | avg log-loss and node count: log-loss ~2% higher than CTW; ~28x fewer nodes | Mixed |
| enforce node bound L=1000 on the tree (D2-CTW uses fringe expansion and pruning, CTW stops growing) | CTW vs D2-CTW · Calgary Corpus | avg log-loss: CTW 11.4% higher on average | Improved |
| dynamic expansion with significance test (D2-CTW) under 5% uniform noise plus node bound | CTW under noise · Calgary Corpus with 5% noise, L=1000 | avg log-loss: CTW 20.0% worse | Improved |
| dynamic depth (D2-CTW) instead of fixed K=3 CTW/PPM-C | model-based RL in T-Maze · T-Maze length 4 | average reward per episode | Improved |
| node bound L=1000 applied to all methods | D2-CTW vs CTW and PPM-C in cheese maze · Cheese Maze | average reward per step | Improved |
| D2-CTW with node bound L=1000 instead of CTW/PPM-C | partially observable mountain car · P.O. Mountain Car, L=1000 | average reward per episode | Improved |
| varying likelihood ratio threshold gamma (ablation) | D2-CTW · Calgary Corpus (book2, 10% noise) | avg log-loss | No clear effect |
Closest papers by meaning
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.86)
- XLand-100B: A Large-Scale Multi-Task Dataset for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Balancing Context Length and Mixing Times for Reinforcement Learning at Scale (NeurIPS 2024 · similarity 0.85)
- Zero-shot Model-based Reinforcement Learning using Large Language Models (ICLR 2025 · similarity 0.85)
- In-Context Reinforcement Learning for Variable Action Spaces (ICML 2024 · similarity 0.85)
- Bridging State and History Representations: Understanding Self-Predictive RL (ICLR 2024 · similarity 0.85)
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.
