SciPost Submission Page
Deep Set Auto Encoders for Anomaly Detection in Particle Physics
by Bryan Ostdiek
This Submission thread is now published as
Submission summary
| Authors (as registered SciPost users): | Bryan Ostdiek |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2109.01695v2 (pdf) |
| Date accepted: | 2021-12-08 |
| Date submitted: | 2021-11-16 17:51 |
| Submitted by: | Ostdiek, Bryan |
| Submitted to: | SciPost Physics |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Experimental, Computational |
Abstract
There is an increased interest in model agnostic search strategies for physics beyond the standard model at the Large Hadron Collider. We introduce a Deep Set Variational Autoencoder and present results on the Dark Machines Anomaly Score Challenge. We find that the method attains the best anomaly detection ability when there is no decoding step for the network, and the anomaly score is based solely on the representation within the encoded latent space. This method was one of the top-performing models in the Dark Machines Challenge, both for the open data sets as well as the blinded data sets.
Author comments upon resubmission
Referee request: Some of the plots are difficult to interpret (specifically Fig. 4 and 5 - suggest combining the rarer backgrounds into one histogram).
Our Response: Thank you for this recommendation. We have changed the figures to include a much smaller subset of the backgrounds, with the remaining ones marked as “Other”
Referee request: Jargon and grammatical errors should be fixed.
Our Response: many terms have been defined and grammar has been fixed. Glad to correct any more specific instances.
Referee request: Since the decoder portion is concluded to be unnecessary for anomaly detection, I suggest either a clear separation between the results with and without β=1 or a discussion of their relative performance on certain signals. I would suggest possibly also expanding the results in Fig. 6 to show the full set of TI values for a set of models (both with and without β =1) since anomaly detection means we do not know which model is out there and thus the distribution across a range of models is more valuable than the min/median/max.
Our Response: We have tried to make the results with and without the decoder more clear. In making Fig. 6 now show the full distribution, some of the trends we previously discussed about the method with the decoder are more challenging to see. Therefore, we removed those discussions, and now the results basically show that the version without the decoder is much better, so then we only discuss that. We still keep all of the information about the decoder and the discussion about it in the methods section, because they could be relevant for related studies, and it was a part of this experiment.
List of changes
Added citation to introduction for similar work.
Figures 4 and 5 now show fewer of the background individually to improve readability.
Figure 6 now shows full distribution rather than min, median, max only. Changed some discussion around the figure.
Published as SciPost Phys. 12, 045 (2022)
Reports on this Submission
Report
Thank you for the revisions and responses. I think the paper reads quite nicely now and the interesting observation you make about the latent space and decoder is now at the forefront of the discussion. I have no other comments at this time.