Topological Obstructions to Autoencoding

Abstract

Autoencoders have been proposed as a powerful tool for model-independent anomaly detection in high-energy physics. The operating principle is that events which do not belong to the space of training data will be reconstructed poorly, thus flagging them as anomalies. We point out that in a variety of examples of interest, the connection between large reconstruction error and anomalies is not so clear. In particular, for data sets with nontrivial topology, there will always be points that erroneously seem anomalous due to global issues. Conversely, neural networks typically have an inductive bias or prior to locally interpolate such that undersampled or rare events may be reconstructed with small error, despite actually being the desired anomalies. Taken together, these facts are in tension with the simple picture of the autoencoder as an anomaly detector. Using a series of illustrative low-dimensional examples, we show explicitly how the intrinsic and extrinsic topology of the dataset affects the behavior of an autoencoder and how this topology is manifested in the latent space representation during training. We ground this analysis in the discussion of a mock "bump hunt" in which the autoencoder fails to identify an anomalous "signal" for reasons tied to the intrinsic topology of $n$-particle phase space.

arXiv comment24 + 20 pages, 26 figures; no autoencoders were harmed in the making of this project. v2: JHEP published version

References in corpus (36)

• Identifying Boosted Objects with N-subjettiness
• Deep learning in color: towards automated quark/gluon jet discrimination
• Jet-Images -- Deep Learning Edition
• Deep-learning Top Taggers or The End of QCD?
• Searching for New Physics with Deep Autoencoders
• Jet Substructure Classification in High-Energy Physics with Deep Neural Networks
• Energy Flow Networks: Deep Sets for Particle Jets
• Equivariant flow-based sampling for lattice gauge theory
• Deep-learned Top Tagging with a Lorentz Layer
• Jet Flavor Classification in High-Energy Physics with Deep Neural Networks
• Jet-Images: Computer Vision Inspired Techniques for Jet Tagging
• A generic anti-QCD jet tagger
• QCD or What?
• Novelty Detection Meets Collider Physics
• Adversarially-trained autoencoders for robust unsupervised new physics searches
• Parton Shower Uncertainties in Jet Substructure Analyses with Deep Neural Networks
• Energy flow polynomials: A complete linear basis for jet substructure
• Pulling Out All the Tops with Computer Vision and Deep Learning
• The Lund Jet Plane
• Variational Autoencoders for New Physics Mining at the Large Hadron Collider
• QCD-Aware Recursive Neural Networks for Jet Physics
• The Metric Space of Collider Events
• Sampling using $SU(N)$ gauge equivariant flows
• How Much Information is in a Jet?
• Recursive Neural Networks in Quark/Gluon Tagging
• Finding New Physics without learning about it: Anomaly Detection as a tool for Searches at Colliders
• Novel Jet Observables from Machine Learning
• Jet Charge and Machine Learning
• The Hidden Geometry of Particle Collisions
• A Theory of Quark vs. Gluon Discrimination
• Linearized Optimal Transport for Collider Events
• Towards Machine Learning Analytics for Jet Substructure
• Reports of My Demise Are Greatly Exaggerated: $N$-subjettiness Taggers Take On Jet Images
• Deep learning for the R-parity violating supersymmetry searches at the LHC
• Cutting Multiparticle Correlators Down to Size
• Covariantizing Phase Space

Cited by in corpus (23)

• The Dark Machines Anomaly Score Challenge: Benchmark Data and Model Independent Event Classification for the Large Hadron Collider
• Classifying Anomalies THrough Outer Density Estimation (CATHODE)
• Autoencoders for unsupervised anomaly detection in high energy physics
• Decoding Photons: Physics in the Latent Space of a BIB-AE Generative Network
• Machine Learning for Anomaly Detection in Particle Physics
• Rare and Different: Anomaly Scores from a combination of likelihood and out-of-distribution models to detect new physics at the LHC
• Challenges for Unsupervised Anomaly Detection in Particle Physics
• Online-compatible Unsupervised Non-resonant Anomaly Detection
• Deep Set Auto Encoders for Anomaly Detection in Particle Physics
• Learning to Simulate High Energy Particle Collisions from Unlabeled Data
• What's Anomalous in LHC Jets?
• Anomaly Detection under Coordinate Transformations
• ELSA -- Enhanced latent spaces for improved collider simulations
• Searching for dark jets with displaced vertices using weakly supervised machine learning
• Hypergraphs in LHC Phenomenology -- The Next Frontier of IRC-Safe Feature Extraction
• Non-resonant Anomaly Detection with Background Extrapolation
• Simulation-based Anomaly Detection for Multileptons at the LHC
• Unsupervised and lightly supervised learning in particle physics
• Revealing the Milky Way's Most Recent Major Merger with a Gaia EDR3 Catalog of Machine-Learned Line-of-Sight Velocities
• Preserving New Physics while Simultaneously Unfolding All Observables
• Dynamical transition in controllable quantum neural networks with large depth
• Creating Simple, Interpretable Anomaly Detectors for New Physics in Jet Substructure
• Graph theory inspired anomaly detection at the LHC