AdvNF: Reducing Mode Collapse in Conditional Normalising Flows using Adversarial Learning
arXiv:2401.15948 · doi:10.21468/SciPostPhys.16.5.132
Abstract
Deep generative models complement Markov-chain-Monte-Carlo methods for efficiently sampling from high-dimensional distributions. Among these methods, explicit generators, such as Normalising Flows (NFs), in combination with the Metropolis Hastings algorithm have been extensively applied to get unbiased samples from target distributions. We systematically study central problems in conditional NFs, such as high variance, mode collapse and data efficiency. We propose adversarial training for NFs to ameliorate these problems. Experiments are conducted with low-dimensional synthetic datasets and XY spin models in two spatial dimensions.
29 pages, submitted to Scipost Physics
References in corpus (27)
- Conditional Generative Adversarial Nets
- Generative Adversarial Networks
- An Introduction to Variational Autoencoders
- Machine learning and the physical sciences
- Machine learning phases of matter
- NIPS 2016 Tutorial: Generative Adversarial Networks
- General state space Markov chains and MCMC algorithms
- Unsupervised learning of phase transitions: from principal component analysis to variational autoencoders
- VEEGAN: Reducing Mode Collapse in GANs using Implicit Variational Learning
- Deep Generative Stochastic Networks Trainable by Backprop
- Identifying topological order through unsupervised machine learning
- Flow-based generative models for Markov chain Monte Carlo in lattice field theory
- Equivariant flow-based sampling for lattice gauge theory
- Machine learning vortices at the Kosterlitz-Thouless transition
- Estimation of Thermodynamic Observables in Lattice Field Theories with Deep Generative Models
- Physics-informed deep generative models
- Evaluating generative models in high energy physics
- Machine Learning of Frustrated Classical Spin Models. II. Kernel Principal Component Analysis
- Flow-based sampling in the lattice Schwinger model at criticality
- Flow-based sampling for multimodal and extended-mode distributions in lattice field theory
- Generative models for sampling and phase transition indication in spin systems
- Machine Learning of Thermodynamic Observables in the Presence of Mode Collapse
- Stochastic Normalizing Flows
- Towards meaningful physics from generative models
- Greedification Operators for Policy Optimization: Investigating Forward and Reverse KL Divergences
- Generative learning for the problem of critical slowing down in lattice Gross Neveu model
- Adaptation of the Independent Metropolis-Hastings Sampler with Normalizing Flow Proposals