Calibrating Bayesian Generative Machine Learning for Bayesiamplification
arXiv:2408.00838 · doi:10.1088/2632-2153/ad9136
Abstract
Recently, combinations of generative and Bayesian machine learning have been introduced in particle physics for both fast detector simulation and inference tasks. These neural networks aim to quantify the uncertainty on the generated distribution originating from limited training statistics. The interpretation of a distribution-wide uncertainty however remains ill-defined. We show a clear scheme for quantifying the calibration of Bayesian generative machine learning models. For a Continuous Normalizing Flow applied to a low-dimensional toy example, we evaluate the calibration of Bayesian uncertainties from either a mean-field Gaussian weight posterior, or Monte Carlo sampling network weights, to gauge their behaviour on unsteady distribution edges. Well calibrated uncertainties can then be used to roughly estimate the number of uncorrelated truth samples that are equivalent to the generated sample and clearly indicate data amplification for smooth features of the distribution.
15 pages, 6 figures, updated references, fixed typo
References in corpus (28)
- Adam: A Method for Stochastic Optimization
- Variational Inference with Normalizing Flows
- Weight Uncertainty in Neural Networks
- Hands-on Bayesian Neural Networks -- a Tutorial for Deep Learning Users
- Stochastic Gradient Hamiltonian Monte Carlo
- A Roadmap for HEP Software and Computing R&D for the 2020s
- Bayesian Neural Networks: An Introduction and Survey
- Machine Learning and LHC Event Generation
- Classifying Anomalies THrough Outer Density Estimation (CATHODE)
- Flow Matching for Generative Modeling
- What Are Bayesian Neural Network Posteriors Really Like?
- Understanding Event-Generation Networks via Uncertainties
- GANplifying Event Samples
- Fast Point Cloud Generation with Diffusion Models in High Energy Physics
- Deep-Learning Jets with Uncertainties and More
- How to Understand Limitations of Generative Networks
- CaloScore v2: Single-shot Calorimeter Shower Simulation with Diffusion Models
- PC-Droid: Faster diffusion and improved quality for particle cloud generation
- The Interplay of Machine Learning--based Resonant Anomaly Detection Methods
- Deep Generative Models for Detector Signature Simulation: A Taxonomic Review
- Jet Diffusion versus JetGPT -- Modern Networks for the LHC
- Combine and Conquer: Event Reconstruction with Bayesian Ensemble Neural Networks
- Bayesian Neural Networks for Fast SUSY Predictions
- EPiC-ly Fast Particle Cloud Generation with Flow-Matching and Diffusion
- Interpretable Uncertainty Quantification in AI for HEP
- CaloGraph: Graph-based diffusion model for fast shower generation in calorimeters with irregular geometry
- Classifier Surrogates: Sharing AI-based Searches with the World
- AdamMCMC: Combining Metropolis Adjusted Langevin with Momentum-based Optimization