Versatile Energy-Based Probabilistic Models for High Energy Physics

As a classical generative modeling approach, energy-based models have the natural advantage of flexibility in the form of the energy function. Recently, energy-based models have achieved great success in modeling high-dimensional data in computer vision and natural language processing. In line with these advancements, we build a multi-purpose energy-based probabilistic model for High Energy Physics events at the Large Hadron Collider. This framework builds on a powerful generative model and describes higher-order inter-particle interactions.It suits different encoding architectures and builds on implicit generation. As for applicational aspects, it can serve as a powerful parameterized event generator for physics simulation, a generic anomalous signal detector free from spurious correlations, and an augmented event classifier for particle identification.

Paper

Similar papers

Peer review

Reviewer bfxx6/10 · confidence 3/52023-06-22

Summary

In this paper, the authors describe an energy-based generative framework for modeling the behavior of elementary particles. This framework can then be used to generate High Energy Physics events, similar to those at LHC, as well as be used for anomaly detection, specifically QCD jets.

Strengths

The strength of the paper lies in the clarity of writing, technical detail, and interesting application of the proposed methodology.

Weaknesses

The main weakness of the paper is that it is hard to gauge how the improvements of the methodology over the baselines would actually translate in improvements of event detection at LHC.

Questions

1. In the paper the line "The results of generative sampling are shown in Appendix C" refers to actually refer to Appendix B.2. Please correct this if it is indeed the case. Also, please describe more the dataset used (for example, does it have class imbalances?) and consider including some confusion matrix or supplementary methods to better compare EBM-CLF to ParticleNet 2. Please include in figure 4 left, the curve for EMB-CLF(E(x)) AUC(Top). 3. Table 2 is a bit hard to read because of the many shades of gray and the inclusion of the background QCD classes. Please reconsider how to show the comparison more clearly.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

Yes.

Reviewer Xiwj7/10 · confidence 4/52023-07-05

Summary

This paper proposes to use EBMs to model the fundamental interactions between elementary particles. The paper first introduces a set of techniques to enable effective learning of energy functions across elementary particles, including the use of a classifier to model the conditional distributions. The paper then proposes as an architecture to capture permutation invariant interaction between particles. The paper proposes a set of metrics to measure the generation quality of EBMs as a substitution for maximum likelihood estimation and illustrate how EBMs can effectively detect anomalies between particles.

Strengths

- The paper tackles an interesting and timely problem and explore how generative models (EBMs) can be used in the scientific application of particle dynamics. - The paper is clearly written and experiments demonstrates the utility of EBMs at both modeling the behaviors of underlying particles and detecting spurious patterns - The introduction compelling introduces and motivates the problem studied by the paper - In comparison to many other generative models, EBMs are much less well studied, and this paper represents an original and interesting application of EBMs to different use cases - The method section has a variety of interesting modifications and tricks over prior work (such as the removal of the entropy term) that are interesting to read

Weaknesses

- It might be interesting to see the extent to which the energy function learned by EBM correlates to real physical energy. - It might also be interesting to structure/augment the energy function with ground truth physical energy functions

Questions

See weaknesses

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

4 excellent

Limitations

Yes

Reviewer PfZ95/10 · confidence 4/52023-07-07

Summary

This work applies recent EBM techniques to model jet streams from the LHC. The work adopts a transformer architecture for the EBM to allow a permutation invariant representation that captures high order relations between particles. EBM models of LHC particles are learned using techniques derived from recent image EBM works. Samples generated from the model are shown to closely match statistics of the observed data, and OOD detection with the learned EBM is performed and shown to have superior performance compared to VAE methods.

Strengths

* This paper is the first to explore EBM learning in the area of high-energy physics. It is exciting and interesting to see research into EBMs for new and challenging domains. * Experimental results provide convincing evidence of the efficacy of the learned model for generation and OOD.

Weaknesses

* The paper lacks theoretical and methodological novelty. The work primarily focuses on applications of EBMs to a new domain. * It appears that the primary practical application of the paper is OOD detection. In this case, it might be more effective to use an EBM designed specifically for OOD, as in [a]. I am not sure use of a generative EBM is well-justified. Tailoring the methodology more towards OOD would increase the focus of the paper, which tends to wander somewhat. [a] https://arxiv.org/pdf/2010.03759.pdf

Questions

* In the view of the authors, what are the most important novel aspects of their work?

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

2 fair

Contribution

2 fair

Limitations

Limitations were not discussed.

Reviewer npbJ6/10 · confidence 3/52023-07-23

Summary

This paper explores the use of energy based models to modeling the distribution of jets in high energy particle physics. Here, the data consists of a vector of events, each of which is described by a momentum transverse to the beam and 2D spatial coordinates. The goal is to learn the true distribution of jets from real data via an energy based model, which can subsequently be used for Out-Of-Distribution detection and classification, in the manner of the paper "Your classifier is secretly an energy based model and you should treat it like one". The model is trained with gradient ascent, using Langevin sampling to estimate the intractable expectation in the loss function, and making use of the (persistent) Contrastive Divergence of the loss. An additional KL term to minimize the difference between the true model distribution and the Langevin approximation is used.

Strengths

The paper appears to achieve success in a difficult domain of application. Their validation suggests that the energy based model they train successfully captures the desired data distribution, at least up to well-known features. Furthermore, they show that the energy based model can be altered to a supervised setting, and used as a classifier. The use of the transformer as the neural architecture to model inter-particle interactions also appears to be a successful choice.

Weaknesses

The ideas in this paper were often conveyed poorly, and it was quite hard to understand the technical details, either of the energy based model training procedure, or of the scientific problem in question. In equation 2, x+ and x- were not clearly explained, making the paper hard to read for someone unfamiliar with the details of contrastive divergence. Using the expectation notation of (in Latex) $E_{x^+ \sim p_D}$ would have helped clarify that the x+ are drawn from the data distribution and the x- from the (approximation of) the model distribution. Algorithm 1 describes the "stopping gradient", but more information is needed here. The equation in Figure 1 are blurry - it needs to be redone in higher resolution. More explanation is needed for the anomaly detection task. I did not understand the set up in adequate detail. You say: "With an energy-based model, which is trained on the QCD background events or directly on the slightly signal-contaminated data, we expect unseen signals to have higher energies and correspondingly lower likelihoods". But what is a signal vs a background event here? This needs to be clear for a non-HEP audience. What the labels are in the classification task is unclear. The paper notes that "A more intriguing property of EBMs is that spurious correlations can be better handled", but it wasn't clear to me either what spurious correlations referred to here, or why EBMs would be better. "More interestingly, compositionality can be easily incorporated within the framework of EBMs by simply summing up the energy functions ": while I understand what is meant here, the word "compositionality" means many things in different contexts, and is too vague. While other papers on EBMs have used "compositionality" in this sense, you should either cite them, or explain directly what you mean by it.

Questions

Which are the original contributions of the paper on the machine learning side, if any? In particular, is the particular version of Contrastive Divergence used, (in the section KL Divergence-Improved EBM Training) key to your success here? I'd like to see more discussion of this KL term, which is quite interesting. I would be interested to know how this paper fits into a larger research project. Are there further directions that could be pursued, in terms of applications? Did the authors find the approach satisfactory, or were there limitations that concerned them? Suggestions: I was able to understand the structure of the paper fully only after reading the following two papers: https://arxiv.org/pdf/1912.03263.pdf and https://arxiv.org/pdf/1903.08689.pdf. The first describes the three applications of energy models you are interested in (generative modeling, OOD detection, and classification). The second describes the Contrastive divergence approach, and cites detailed notes which explain its derivation; your training algorithm appears to be a modification of their Algorithm 1. I would summarize your paper as "an application of the approach of these papers to HEP Jet modeling". While you do cite both of these papers, I would recommend highlighting them more, since they form the backbone of your approach and will help the reader to follow along.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

2 fair

Contribution

3 good

Limitations

Are there ways of incorporating domain specific models of HEP into the current approach? Currently it seems that the physics incorporated into simulators is not leveraged here. It would also be interesting if known expressions or heuristics for the energy could be incorporated into the energy function.

Reviewer bfxx2023-08-10

Response following rebuttal

Dear authors, Thank you for the time taken to write the rebuttal and addressing the points I raised.

Reviewer PfZ92023-08-17

Thanks for the author responses. I will keep my score for now.

Thanks to the authors for their responses to myself and other reviewers. I find the work done by the authors to be very interesting and relevant from the HEP perspective. However, I still find the methodology to be essentially the same as commonly used EBM methodology in the image domain. Thus, I am not sure whether this work will have a high degree of impact from the machine learning perspective, and I choose to keep my score for now. I would be willing to reconsider my score if there is a clear contribution to EBM learning beyond what is commonly performed in the image domain.

Authorsrebuttal2023-08-18

Though we acknowledge the reviewer's perspective, we emphasize that this work is submitted under the “Machine Learning for Physical Sciences” track. This work's focus and primary objective is to solve scientific problems with advanced machine learning techniques. AI for Science projects often transcend methodological novelty, as seen in AlphaFold's usage of the established attention mechanism, without diminishing its scientific impact. NeurIPS is a general venue that encourages interdisciplinary submissions. Limiting NeurIPS to methodological novelty could narrow its scientific vision and hinder interdisciplinary scientific communication. We thank the reviewer for their understanding.

Reviewer PfZ92023-08-18

Thanks for pointing out that this paper is submitted in the area of Machine Learning for Physical Sciences, I overlooked this in my previous comments. Given the positive feedback from other reviewers and the fresh application perspective in this work, I am willing to raise my score to borderline accept. I still believe the work has a modest degree of novelty, most of which comes from the area of application. Nonetheless, the direction taken by the authors is interesting and I look forward to seeing more application of EBM and other generative models in the physical sciences in future works.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC