SciPost logo

SciPost Submission Page

Enhancing low energy reconstruction and classification in KM3NeT/ORCA with transformers

by Iván Mozún Mateo

Submission summary

Authors (as registered SciPost users): Iván Mozún Mateo
Submission information
Preprint Link: https://arxiv.org/abs/2511.18999v1  (pdf)
Date submitted: Dec. 1, 2025, 1:40 p.m.
Submitted by: Iván Mozún Mateo
Submitted to: SciPost Physics Proceedings
Proceedings issue: The 2nd European AI for Fundamental Physics Conference (EuCAIFCon2025)
Ontological classification
Academic field: Physics
Specialties:
  • High-Energy Physics - Experiment
Approaches: Experimental, Computational

Abstract

The current KM3NeT/ORCA neutrino telescope, still under construction, has not yet reached its full potential in neutrino reconstruction capability. When training any deep learning model, no explicit information about the physics or the detector is provided, thus they remain unknown to the model. This study leverages the strengths of transformers by incorporating attention masks inspired by the physics and detector design, making the model understand both the telescope design and the neutrino physics measured on it. The study also shows the efficacy of transformers on retaining valuable information between detectors when doing fine-tuning from one configurations to another.

Current status:
Awaiting resubmission

Reports on this Submission

Report #1 by Thomas Vuillaume (Referee 1) on 2025-12-5 (Invited Report)

Disclosure of Generative AI use

The referee discloses that the following generative AI tools have been used in the preparation of this report:

chatGPT 5.1, 05/12/2025 - spelling and prephrasing

Strengths

  1. Physics-informed attention masks. This is an interesting idea to inject physics into the transformer architecture.
  2. Transfer learning across configurations. This is highly relevant to KM3NeT but also to other experiments with different and/or growing configurations.

Weaknesses

  1. no study of the impact of the PairwiseAttention compared to standard Attention. Although the idea seems interesting, experimental proof of its validity and impact would be expected.

Report

The manuscript presents a transformer-based approach for the event reconstruction and classification in KM3NeT/ORCA. It is tested on simulated data and studies the performance improvement when trained on a larger telescope configuration and fine-tuned on a smaller one.
I recommended it for publication with minor changes.

Requested changes

  1. "PMTs detects light" p.2 -> "PMTs detect light"
  2. "it computes the score it from what the raw data", p.3 - rephrase
  3. "These techniques ... must be updated whenever a new version of the reconstruction software is released.", p.3 - I suppose the approach proposed here still requires a calibration step (not discussed in the paper) which also makes it dependent on the reconstruction software?
  4. "A advantage of transformers ", p.4 -> "An .."
  5. " the inference time is significantly reduced" - compared to what?

Recommendation

Ask for minor revision

  • validity: top
  • significance: high
  • originality: high
  • clarity: high
  • formatting: good
  • grammar: good

Login to report

Comments

Anonymous on 2025-12-04  [id 6105]

Category:
remark

Resubmission properly as an arXiv