Research map / Large language models / Agents, code and math
TusoAI: Agentic Optimization for Scientific Methods
Alistair Turcan, Kexin Huang, Lei Li, Martin Jinye Zhang
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 |
|---|---|---|---|
| add domain knowledge (knowledge tree from papers) (ablation) | TusoAI · ablation set average | average score: 0.46 -> 0.66 | Improved |
| add category structure for instructions (ablation) | TusoAI · ablation set average | average score: 0.57 -> 0.66 | Improved |
| add Bayesian category sampling (ablation) | TusoAI · ablation set average | average score: 0.55 -> 0.66 | Improved |
| add model diagnosis step (ablation) | TusoAI · ablation set average | average score: 0.64 -> 0.66 | Improved |
| use GPT-5 instead of GPT-4o-mini backbone (ablation) | TusoAI · ablation set average | average score: 0.66 -> 0.69 | No clear effect |
| use Claude-4-Sonnet instead of Claude-3.5-Haiku backbone (ablation) | TusoAI · ablation set average | average score: 0.68 -> 0.61 | Hurt |
| use open-source GPT-oss-120b backbone (ablation) | TusoAI · ablation set average | average score: 0.66 -> 0.54 | Hurt |
| use TusoAI instead of AIDE | single-cell tasks · Single-cell benchmarks | average score: 0.52 -> 0.66 | Improved |
| use TusoAI instead of DARTS | NASBench-360 tasks · NASBench-360 average | average score: 0.71 -> 0.70 | Hurt |
| optimize scDRS scoring function with TusoAI | scDRS · real-data cell type-disease associations | true associations found: 14 -> 17 | Improved |
| optimize pgBoost distance features with TusoAI | pgBoost · fine-mapped eQTL links | enrichment: +13.8% enrichment (eQTL), +7.2% (ABC) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Expert — ExPERT: Modeling Human Behavior Under External Stimuli Aware Personalized MTPP (AAAI 2025) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- InnovatorBench: Evaluating Agents’ Ability to Conduct Innovative AI Research (ICLR 2026 · similarity 0.88)
- AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite (ICLR 2026 · similarity 0.88)
- ScienceBoard: Evaluating Multimodal Autonomous Agents in Realistic Scientific Workflows (ICLR 2026 · similarity 0.88)
- BioDiscoveryAgent: An AI Agent for Designing Genetic Perturbation Experiments (ICLR 2025 · similarity 0.87)
- SciAgentGym: Benchmarking Multi-Step Scientific Tool-Use in LLM Agents (ICML 2026 · similarity 0.87)
- SciResearchBench: Benchmarking AI Agents on Complex Scientific Literature Discovery (NeurIPS 2026 · 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.
