Explainable AI for ML jet taggers using expert variables and layerwise relevance propagation
arXiv:2011.13466 · doi:10.1007/JHEP05(2021)208
Abstract
A framework is presented to extract and understand decision-making information from a deep neural network (DNN) classifier of jet substructure tagging techniques. The general method studied is to provide expert variables that augment inputs ("eXpert AUGmented" variables, or XAUG variables), then apply layerwise relevance propagation (LRP) to networks both with and without XAUG variables. The XAUG variables are concatenated with the intermediate layers after network-specific operations (such as convolution or recurrence), and used in the final layers of the network. The results of comparing networks with and without the addition of XAUG variables show that XAUG variables can be used to interpret classifier behavior, increase discrimination ability when combined with low-level features, and in some cases capture the behavior of the classifier completely. The LRP technique can be used to find relevant information the network is using, and when combined with the XAUG variables, can be used to rank features, allowing one to find a reduced set of features that capture part of the network performance. In the studies presented, adding XAUG variables to low-level DNNs increased the efficiency of classifiers by as much as 30-40\%. In addition to performance improvements, an approach to quantify numerical uncertainties in the training of these DNNs is presented.
arXiv comment38 pages, 30 figures
References in corpus (35)
- The anti-k_t jet clustering algorithm
- An Introduction to PYTHIA 8.2
- FastJet user manual
- Interpretable machine learning: definitions, methods, and applications
- On the Convergence of Adam and Beyond
- Explaining NonLinear Classification Decisions with Deep Taylor Decomposition
- Identifying Boosted Objects with N-subjettiness
- Soft Drop
- How to Explain Individual Classification Decisions
- Towards an understanding of jet substructure
- Towards Jetography
- Performance of -Jet Identification in the ATLAS Experiment
- Maximizing Boosted Top Identification by Minimizing N-subjettiness
- Boosted objects: a probe of beyond the Standard Model physics
- Jet Substructure at the Large Hadron Collider: A Review of Recent Advances in Theory and Machine Learning
- Jet Substructure at the Tevatron and LHC: New results, new tools, new benchmarks
- Jet Substructure at the Large Hadron Collider: Experimental Review
- Jet-Images -- Deep Learning Edition
- Deep-learning Top Taggers or The End of QCD?
- Boosted objects and jet substructure at the LHC
- Jet Flavor Classification in High-Energy Physics with Deep Neural Networks
- Seeing in Color: Jet Superstructure
- Jet-Images: Computer Vision Inspired Techniques for Jet Tagging
- How to GAN LHC Events
- Decorrelated Jet Substructure Tagging using Adversarial Neural Networks
- JEDI-net: a jet identification algorithm based on interaction networks
- ABCNet: An attention-based method for particle tagging
- Thinking outside the ROCs: Designing Decorrelated Taggers (DDT) for jet substructure
- How Much Information is in a Jet?
- Interaction networks for the identification of boosted decays
- Mapping Machine-Learned Physics into a Human-Readable Space
- Interpretable Deep Learning for Two-Prong Jet Classification with Jet Spectra
- Towards Machine Learning Analytics for Jet Substructure
- Spectral Analysis of Jet Substructure with Neural Networks: Boosted Higgs Case
- Deep Learning Jet Substructure from Two-Particle Correlation