AI Paper Atlas

Research map / Large language models / Agents, code and math

DeepAnalyze: Agentic Large Language Models for Autonomous Data Science

Shaolei Zhang, Ju Fan, Meihao Fan, Yizhe Liu, Yuxin Zhang, Xiaoyong Du

ICML 2026 · Large language models › Agents, code and math · approach: Long-horizon LLM agents · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
add the separate <Understand> action (ablation)DeepAnalyze-8B · DABStepAccuracy: 31.78 -> 38.88Improved
add the separate <Understand> action (ablation)DeepAnalyze-8B · WikiTQExact match: 80.78 -> 83.24Improved
add the separate <Understand> action (ablation)DeepAnalyze-8B · DS-1000Pass@1: 61.20 -> 61.70No clear effect
add multi-ability agentic training after single-ability fine-tuning (ablation)DeepAnalyze-8B · DABStepAccuracy: 15.34 -> 38.88Improved
add single-ability fine-tuning before agentic training (ablation)DeepAnalyze-8B · DABStepAccuracy: 30.66 -> 38.88Improved
use two-stage curriculum instead of one-stage mixed training (ablation)DeepAnalyze-8B · DS-1000Pass@1: 54.80 -> 61.70Improved
use two-stage curriculum instead of one-stage mixed training (ablation)DeepAnalyze-8B · DABStepAccuracy: 36.89 -> 38.88Improved
add distillation of reasoning trajectories (ablation)Reasoning-Table data, trained LLM · WikiTQExact match: 75.54 -> 78.80Improved
add keyword-guided refinement on top of distillation (ablation)Reasoning-Table data, trained LLM · WikiTQExact match: 78.80 -> 80.25Improved
add keyword-guided refinement on top of distillation (ablation)Reasoning-Table data, trained LLM · HybridQAExact match: 36.12 -> 38.84Improved
add keyword-guided refinement on top of distillation (ablation)Reasoning-Table data, trained LLM · HiTabExact match: 74.44 -> 75.86Improved
add agentic training stage on top of single-ability fine-tuning (ablation)DeepAnalyze-8B · TableQA (7-benchmark average)Exact match (AVG): 60.64 -> 64.47Improved

Later papers that compared with this method

1 later accepted paper reports DeepAnalyze as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy