SciPost Submission Page
Scaling Laws in Jet Classification
by Joshua Batson, Yonatan Kahn
This Submission thread is now published as
Submission summary
| Authors (as registered SciPost users): | Yonatan Frederick Kahn |
| Submission information | |
|---|---|
| Preprint Link: | scipost_202412_00008v2 (pdf) |
| Date accepted: | 2025-03-19 |
| Date submitted: | 2025-02-21 17:52 |
| Submitted by: | Kahn, Yonatan Frederick |
| Submitted to: | SciPost Physics Core |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Theoretical, Computational, Phenomenological |
Abstract
We demonstrate the emergence of scaling laws in the benchmark top versus QCD jet classification problem in collider physics. Six distinct physically-motivated classifiers exhibit power-law scaling of the binary cross-entropy test loss as a function of training set size, with distinct power law indices. This result highlights the importance of comparing classifiers as a function of dataset size rather than for a fixed training set, as the optimal classifier may change considerably as the dataset is scaled up. We speculate on the interpretation of our results in terms of previous models of scaling laws observed in natural language and image datasets.
Author comments upon resubmission
List of changes
1. Following comments by Referees 1 and 2, we have added a paragraph on p. 10 (lines 313-323) summarizing the results of Ref. [4] and giving context for the data-data covariance matrix.
2. Following comments by Referee 1, we have added a sentence on p. 12 (lines 368-372) speculating on the poorer fit for including a nonzero loss floor in the non-DNN classifiers.
Published as SciPost Phys. Core 8, 034 (2025)