Mathematical & physical sciences · OECD FoS 1.1 · China MoE 0714 统计学
Statistics
统计学
L2Local breakthroughs → Steady Score 1.90 · Last reviewed 2026-10-22 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2025-01-08 AI · majorField · major
TabPFN: tabular foundation model for small data (Nature) ↗
Beats tuned gradient boosting on small tabular data
Why this level
Tabular foundation model TabPFN beats tuned gradient boosting on small tabular data (Nature 2025); prediction-powered inference enables valid inference with ML predictions (Science 2023). Core methods are not yet systematically rewritten.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
Tabular foundation modelsPrediction-powered inferenceData-analysis agents
Papers
- MR-MAS: an end-to-end multi-agent system for autonomous Mendelian randomization analysis
AI: a structured MR knowledge base guides implementation and code repair. Field: reproduces literature-supported causal signals with stable estimates — a peer-reviewed case of agent-automated causal-inference workflows
2026-10-20 · Bioinformatics
- The more you automate, the less you see: Hidden pitfalls of autonomous AI scientists
AI: controlled experiments isolate four pitfalls in two open-source systems. Field: brings leakage, selection bias and selective reporting into automated research and argues for auditing full workflow artefacts
2026-10-05 · PNAS
- Pragmatic DML with AI-Learned Representations
AI: foundation-model embeddings are now routinely used as controls. Field: shows distortion equals a product of two representation errors and gives a practical DML → interpretation → sensitivity workflow (preprint)
2026-10-01 · arXiv
- PPI++: Efficient prediction-powered inference
AI: combines abundant AI predictions with scarce gold labels for valid intervals; Field: prediction-powered inference enters a mainstream statistics journal
2026-09-01 · The Annals of Applied Statistics
- A Survey on Large Language Model-based Agents for Statistics and Data Science
Mainstream stats journal maps the data-agent field
2025-10-16 · The American Statistician
- LAMBDA: A Large Model Based Data Agent
First flagship-journal debate on fully automated data analysis
2025-07-17 · Journal of the American Statistical Association
- Accurate predictions on small data with a tabular foundation model
Beats tuned gradient boosting on small tabular data
2025-01-08 · Nature
- Cross-prediction-powered inference
Extends PPI to settings without a separate labelled model
2024-04-03 · Proceedings of the National Academy of Sciences
- Prediction-powered inference
General framework for valid inference with ML predictions
2023-11-10 · Science
- Beyond Scalar Leaderboards: Adaptive Sampling for Pareto Frontier Identification in Multi-Objective LLM Arenas
VB-EGE adaptively samples arena comparisons to certify the Pareto frontier of LLMs with minimax-optimal sample complexity.
2026-10-20 · arXiv · Liang Twist Shan, Jiahao Ai, Zhimei Ren
- Keep the Lineages Alive: Diversity-Preserving SMC for Diffusion LLM Decoding
Improves the resampling step of inference-time SMC for diffusion LLMs, keeping diversity while focusing compute on promising candidates.
2026-10-19 · arXiv · Minoo Ahmadi, Seyedarmin Azizi, Mehdi Kamal, Massoud Pedram
- Token-Level Attribution of Multi-Mode LLM Watermarks: Optimal Rule and Belief Weighting
Token-level attribution for multi-key Gumbel-max LLM watermarks that controls Type I error while keeping power.
2026-10-16 · arXiv · Rui Zhou, Licheng Zhao, Chao Shen, Wenqiang Pu
- Chance-Constrained Personalized LLM Routing with Online Preference Learning
Formulates personalised LLM routing as a constrained contextual bandit that learns preferences while cutting cost and keeping quality.
2026-10-15 · arXiv · Ameya Anjarlekar, Rasoul Etesami, R Srikant
- Covariance-Calibrated Prediction-Powered Inference
Uses unlabeled prediction scores to calibrate the covariance weight in prediction-powered GMM inference, with spectral protection for positive definiteness.
2026-10-15 · arXiv · Siyuan Chen, Yang Bai, Huichen Zhu
- Learning Checklists to Reduce Lottery Effects in Scientific Peer Review
Learns hierarchical checklists of criticisms from historical reviews to reduce “reviewer lottery” error in peer review.
2026-10-14 · arXiv · Xuanyou Liu, Ziyang Guo, Jason Hartline, Jessica Hullman
- How Low Can You Go: Fine-Tuning LLMs with Bayesian Optimization (LlMBO)
Replaces gradient-based local search with Bayesian optimisation for TinyLoRA fine-tuning with very few parameters.
2026-10-13 · arXiv · Aditya Kulkarni, Davide Sartor, Carlos Soto, Nathan Wycoff
- textGrain watermarking is harder to detect: can sum-tests still do it optimally?
Studies whether sum-type tests remain optimal for detecting the entropy-calibrated textGrain LLM watermark.
2026-10-13 · arXiv · Swapnaneel Bhattacharyya, Soham Bonnerjee
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
Milestones 10
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
System
MR-MAS (Bioinformatics): an LLM multi-agent system that runs end-to-end Mendelian randomization — data retrieval, causal estimation, interpretation and write-upAI: a structured MR knowledge base guides implementation and code repair. Field: reproduces literature-supported causal signals with stable estimates — a peer-reviewed case of agent-automated causal-inference workflows
Bioinformatics -
Paper
PNAS: “The more you automate, the less you see” — open-source AI-scientist systems show benchmark misselection, data leakage, metric misuse and post hoc selection bias that are invisible in the final manuscriptAI: controlled experiments isolate four pitfalls in two open-source systems. Field: brings leakage, selection bias and selective reporting into automated research and argues for auditing full workflow artefacts
PNAS -
Paper
Chernozhukov, Newey et al.: “Pragmatic DML with AI-Learned Representations” — valid DML inference and sensitivity bounds when AI embeddings serve as controlsAI: foundation-model embeddings are now routinely used as controls. Field: shows distortion equals a product of two representation errors and gives a practical DML → interpretation → sensitivity workflow (preprint)
arXivPreprint -
Paper
PPI++: Efficient prediction-powered inference (Ann. Appl. Stat.)AI: combines abundant AI predictions with scarce gold labels for valid intervals; Field: prediction-powered inference enters a mainstream statistics journal
The Annals of Applied Statistics -
Industry
JASA 121(553) publishes LAMBDA with all invited discussionsMarks the statistics community formally debating AI’s impact
ASA Connect -
Paper
The American Statistician survey of LLM agents for statistics & data scienceMainstream stats journal maps the data-agent field
The American Statistician -
Paper
LAMBDA: LLM data agent, JASA discussion paper with invited discussants incl. Donoho, Xihong Lin, Xiao-Li MengFirst flagship-journal debate on fully automated data analysis
JASA -
Paper
TabPFN: tabular foundation model for small data (Nature)Beats tuned gradient boosting on small tabular data
Nature -
Paper
Cross-prediction-powered inference (PNAS)Extends PPI to settings without a separate labelled model
PNAS -
Paper
Prediction-powered inference published in ScienceGeneral framework for valid inference with ML predictions
Science
No milestones match this filter.
More papers & repos 7
| Type | Title | Date |
|---|---|---|
| Paper | PPI++: Efficient prediction-powered inference | 2026-09-01 |
| Paper | A Survey on Large Language Model-based Agents for Statistics and Data Science | 2025-10-16 |
| Paper | LAMBDA: A Large Model Based Data Agent | 2025-07-17 |
| Paper | Accurate predictions on small data with a tabular foundation model | 2025-01-08 |
| Paper | Cross-prediction-powered inference | 2024-04-03 |
| Paper | Prediction-powered inference | 2023-11-10 |
| Repo | PriorLabs/TabPFN To be verified | — |
Latest news 5
- The statistical science behind trustworthy AIGeorge Mason University Statisticsen
- ONS surveys and economic statistics improvement plan, quarterly progress update: October 2026Office for National StatisticsenTo be verified