Research map / Large language models / Agents, code and math
Bonsai: Interpretable Tree-Adaptive Grounded Reasoning
Kate Sanders, Benjamin Van Durme
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 |
|---|---|---|---|
| re-sample visual evidence using leaf sub-claims (test-time evidence scaling) (ablation) | Bonsai · MultiVENT | accuracy: 76 -> 89 | Improved |
| replace mean/product leaf aggregation with GPT-4o judge (ablation) | Bonsai · EntailmentBank | accuracy: 87.7 -> 95.6 | Improved |
| replace mean leaf aggregation with GPT-4o judge (ablation) | Bonsai · TVQA | accuracy: 65.5 -> 68.8 | Improved |
| replace mean aggregation with GPT-4o judge (ablation) | Bonsai · EntailmentBank test split | accuracy: 86.4 -> 94.7 | Improved |
| replace mean aggregation with GPT-4o judge (ablation) | Bonsai · EntailmentBank train split | accuracy: 87.9 -> 96.1 | Improved |
| use Bonsai transparent reasoning instead of end-to-end video models | Bonsai vs TV-TREES · TVQA | accuracy: 49.4 -> 65.5 | Improved |
| use Bonsai instead of transparent TreeWise | Bonsai vs TreeWise · EntailmentBank | accuracy: 79.2 -> 87.7 | Improved |
| use Bonsai without evidence re-sampling instead of Qwen VL 2-7B | Bonsai vs Qwen VL 2-7B · MultiVENT | accuracy: 87 -> 76 | Hurt |
| add anchor-and-adjust evidence scoring versus basic likelihood prompt (ablation) | Bonsai · SQUID-E | agreement with human judgments (MSE) | Improved |
Closest papers by meaning
- Deep Reasoning in General Purpose Agents via Structured Meta-Cognition (NeurIPS 2026 · similarity 0.87)
- Decompose, Analyze and Rethink: Solving Intricate Problems with Human-like Reasoning Cycle (NeurIPS 2024 · similarity 0.86)
- A$^2$FM: An Adaptive Agent Foundation Model for Tool-Aware Hybrid Reasoning (ICLR 2026 · similarity 0.86)
- Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilities (AAAI 2026 · similarity 0.86)
- LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning (NeurIPS 2026 · similarity 0.86)
- RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models (ICLR 2026 · 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.
