iml-wg/HEPML-LivingReview

What it solves

Machine learning is being applied across particle physics at a rapid pace, but the field lacks a single trusted map of the relevant literature. Researchers who want to apply deep learning to experimental, phenomenological, or theoretical analyses face an overwhelming number of papers scattered across journals, preprints, and conference proceedings. This project solves that by providing a nearly comprehensive, curated list of citations, grouped into a small set of practical categories, so practitioners can quickly find the work most relevant to them.

How it works

The review is a living document that was maintained by the HEP-ML community. Papers are collected from the INSPIRE literature database, then organised into a small number of topics to keep the list useful. Each paper is listed with a link to its arXiv or DOI entry, and may appear in more than one category if it spans multiple topics. The classification was done by the community, not by an automated process. The project is now frozen as of 1 June 2026, and the final archival version is available on Zenodo. The PDF and source files are freely accessible, and users are encouraged to cite the archived version using the provided BibTeX key.

Who it's for

This resource is for anyone developing or applying machine learning techniques in high energy physics, including experimentalists, phenomenologists, and theorists. It is also useful for educators, students, and researchers entering the field who need a comprehensive starting point. The review includes classical papers, modern reviews, specialised reviews, datasets, and papers grouped by technical approach, so it serves both newcomers and experienced practitioners looking for a specific method or application.

Highlights

• Nearly comprehensive literature coverage, frozen at June 2026, capturing the state of the field at that time.
• Papers are grouped into a small set of clear categories, including reviews, datasets, parameterised classifiers, jet images, and many other specialised topics.
• Built with community involvement and the INSPIRE REST API, making the curation collaborative and transparent.
• Freely accessible PDF and GitHub repository, with a permanent archival version on Zenodo for stable citation.
• Includes both modern and classical papers, from early neural-network packages to cutting-edge deep learning and foundation models.
• Papers are not endorsed by the review; the classification is a best-effort organisation that leaves validation to peer review and the community.