Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating Physics Inspired Features with Graph Convolution

Rameswar Sahu

TL;DR

The paper tackles quark–gluon tagging with graph neural networks by introducing CapsLorentzNet, which replaces the decoding block of LorentzNet with a Capsule block that uses vector-valued activations and reconstruction-based regularization to incorporate High-Level Features. The Capsule block enables the network to learn and regularize high-level jet features, improving discriminative performance. Empirically, CapsLorentzNet achieves about a $20\%$ gain in background rejection at $50\%$ signal efficiency on both generated and public datasets, though it trades off Lorentz equivariance due to HLFs. The approach is generalizable to other GNNs and provides a practical route to enhance boosted object tagging by integrating expert-designed features through capsule-based decoding and reconstruction.

Abstract

With the advent of advanced machine learning techniques, boosted object tagging has witnessed significant progress. In this article, we take this field further by introducing novel architectural modifications compatible with a wide array of Graph Neural Network (GNN) architectures. Our approach advocates for integrating capsule layers, replacing the conventional decoding blocks in standard GNNs. These capsules are a group of neurons with vector activations. The orientation of these vectors represents important properties of the objects under study, with their magnitude characterizing whether the object under study belongs to the class represented by the capsule. Moreover, capsule networks incorporate a regularization by reconstruction mechanism, facilitating the seamless integration of expert-designed high-level features into the analysis. We have studied the usefulness of our architecture with the LorentzNet architecture for quark-gluon tagging. Here, we have replaced the decoding block of LorentzNet with a capsulated decoding block and have called the resulting architecture CapsLorentzNet. Our new architecture can enhance the performance of LorentzNet by 20 \% for the quark-gluon tagging task.

Integrating Physics Inspired Features with Graph Convolution

TL;DR

The paper tackles quark–gluon tagging with graph neural networks by introducing CapsLorentzNet, which replaces the decoding block of LorentzNet with a Capsule block that uses vector-valued activations and reconstruction-based regularization to incorporate High-Level Features. The Capsule block enables the network to learn and regularize high-level jet features, improving discriminative performance. Empirically, CapsLorentzNet achieves about a gain in background rejection at signal efficiency on both generated and public datasets, though it trades off Lorentz equivariance due to HLFs. The approach is generalizable to other GNNs and provides a practical route to enhance boosted object tagging by integrating expert-designed features through capsule-based decoding and reconstruction.

Abstract

With the advent of advanced machine learning techniques, boosted object tagging has witnessed significant progress. In this article, we take this field further by introducing novel architectural modifications compatible with a wide array of Graph Neural Network (GNN) architectures. Our approach advocates for integrating capsule layers, replacing the conventional decoding blocks in standard GNNs. These capsules are a group of neurons with vector activations. The orientation of these vectors represents important properties of the objects under study, with their magnitude characterizing whether the object under study belongs to the class represented by the capsule. Moreover, capsule networks incorporate a regularization by reconstruction mechanism, facilitating the seamless integration of expert-designed high-level features into the analysis. We have studied the usefulness of our architecture with the LorentzNet architecture for quark-gluon tagging. Here, we have replaced the decoding block of LorentzNet with a capsulated decoding block and have called the resulting architecture CapsLorentzNet. Our new architecture can enhance the performance of LorentzNet by 20 \% for the quark-gluon tagging task.
Paper Structure (7 sections, 6 equations, 3 figures, 3 tables)

This paper contains 7 sections, 6 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Dataset
  3. Model
  4. The LorentzNet block
  5. The Capsule block
  6. Classifier Performance
  7. Summary and Outlook

Figures (3)

  • Figure 1: ROC curves of the LorentzNet classifier trained with the publicly available dataset komiske_2019_3164691 (blue) and our generated dataset (red).
  • Figure 2: ROC curves of the LorentzNet classifier without jet-level information (cyan) and the two classifiers with additional global information (red and blue).
  • Figure 3: ROC curves of the LorentzNet (red) and CapsLorentzNet (blue) Classifiers. The left plot depicts the ROC curves for the models when both training and testing are performed on our dataset. Conversely, the right plot illustrates the performance of the models trained on our dataset but evaluated on a public dataset. The results are averaged over six runs with different random seed initializations.