SciPost Submission Page
Unsupervised Machine Learning for Anomaly Detection in LHC Collider Searches
by Antonio D'Avanzo
Submission summary
| Authors (as registered SciPost users): | Antonio D'Avanzo |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.24723v1 (pdf) |
| Date submitted: | Sept. 30, 2025, 10:11 a.m. |
| Submitted by: | Antonio D'Avanzo |
| Submitted to: | SciPost Physics Proceedings |
| Proceedings issue: | The 2nd European AI for Fundamental Physics Conference (EuCAIFCon2025) |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Experimental, Computational |
Abstract
Searches for new physics at the LHC at CERN traditionally use advanced simulations to model Standard Model and new-physics processes in high-energy collisions and compare them with data. The lack of recent direct discoveries, however, has motivated the development of model-independent approaches in HEP to complement existing hypothesis-driven analyses, particularly Anomaly Detection. A review of the latest efforts in BSM searches with anomaly detection is presented in these proceedings, focusing on contributions within the ATLAS collaboration at LHC and discussing Variational Recurrent Neural Network, Deep Transformer and Graph Anomaly Detection applications.
Current status:
Reports on this Submission
Strengths
1- promising topic, with significant interest among the experimental collaborations
Weaknesses
1- language needs improving 2- the paper could focus more on the study on LHCOlympics dataset and less on presentation of GNN/transformers (Section 3)
Report
Requested changes
1- Section 1, second line : "it's" -> "it is" (throughout) 2- Section 2, first line: "$\approx TeV$" -> consider mentioning it is the mass that is in the TeV range 3- Section 2 : "jet, i.e. the collection of calorimeter energy deposits" : jets in ATLAS are usually made from combination of tracks & calorimeter deposits (particle flow approach). Consider clarifying/dropping that part. 4- Section 3 : missing full stop in 3rd sentence 5- " Graph embedding, if needed, can be at trained ultimated computed by [...]" : consider rephrasing confusing sentence 6- "finding the max" : replace "max" by "maximum" 7- Consider shortening and clarifying Section 3, focusing more on Section 4 (the application) 8- Section 4 : avoid "in fact" 9- Section 4 : Clarify "using as features the fraction transverse momentum". Maybe replace with "fractional transverse momentum" ? 10- Consider explaining the "data augmentation procedure" to decorrelate the jet mass 11- "Constituents features also undergo a data augmentation procedure in order to decorrelate the jets mass from a GNN, specifically an Edge-Featured Graph Attention Network (EGAT) model, or a Transformer prediction, which are the two architecturestested in the R&D phase for now. " : split this sentence into two. 12- "MSE minimization objective in the Transformer case" : consider clarifying what MSE is referring to in unsupervised learning 13- Section 4.1 : "with an assumes" -> "assumed" ? 14- "by the group" : clarify which group is referred to (the ATLAS collaboration ?) 15- references need formatting 16- Figure legend need formatting (subfigures label "(a)" is present but not "(b)")
Recommendation
Ask for minor revision