The sampling of probability distributions specified up to a normalization constant is an important problem in both machine learning and statistical mechanics. While classical stochastic sampling methods such as Markov Chain Monte Carlo (MCMC) or Langevin Dynamics (LD) can suffer from slow mixing times there is a growing interest in using normalizing flows in order to learn the transformation of a simple prior distribution to the given target distribution. Here we propose a generalized and combined approach to sample target densities: Stochastic Normalizing Flows (SNF) - an arbitrary sequence of deterministic invertible functions and stochastic sampling blocks. We show that stochasticity overcomes expressivity limitations of normalizing flows resulting from the invertibility constraint, whereas trainable transformations between sampling steps improve efficiency of pure MCMC/LD along the flow. By invoking ideas from non-equilibrium statistical mechanics we derive an efficient training procedure by which both the sampler's and the flow's parameters can be optimized end-to-end, and by which we can compute exact importance weights without having to marginalize out the randomness of the stochastic blocks. We illustrate the representational power, sampling efficiency and asymptotic correctness of SNFs on several benchmarks including applications to sampling molecular systems in equilibrium.

By invoking ideas from non-equilibrium statistical mechanics we derive an efficient training procedure by which both the sampler's and the flow's parameters can be optimized end-to-end, and by which we can compute

Additional Feedback: The abstract is a bit long and could probably be condensed, and would probably benefit from doing so. It might also be worthwhile to separate the title from the paragraph text rather than joining them as in e.g. Why not make the base distribution pZ? That is, pZ - pX under F. 50-52: Although the slash is being used to distinguish between two different cases, it's ambiguous because the terms could also be interpreted as the ratio between two KL divergences, as well as the ratio between two densities. On relating statistical physics to more classic ML: as you've promised, it would be nice to include a latent variable/variational bound interpretation (as Sohl-Dickstein et al 2015 'Deep Unsupervised Learning using Nonequilibrium Thermodynamics' do), and maybe also link to Deep Latent Gaussian Models (Rezende et al 2014 'Stochastic Backprop', Kingma et al, 'Autoencoding Variational Bayes').

Effective String Theory (EST) is a powerful tool used to study confinement in pure gauge theories by modeling the confining flux tube connecting a static quark-anti-quark pair as a thin vibrating string. Recently, flow-based samplers have been applied as an efficient numerical method to study EST regularized on the lattice, opening the route to study observables previously inaccessible to standard analytical methods. Flow-based samplers are a class of algorithms based on Normalizing Flows (NFs), deep generative models recently proposed as a promising alternative to traditional Markov Chain Monte Carlo methods in lattice field theory calculations. By combining NF layers with out-of-equilibrium stochastic updates, we obtain Stochastic Normalizing Flows (SNFs), a scalable class of machine learning algorithms that can be explained in terms of stochastic thermodynamics. In this contribution, we outline EST and SNFs, and report some numerical results for the shape of the flux tube.

The sampling of probability distributions specified up to a normalization constant is an important problem in both machine learning and statistical mechanics. While classical stochastic sampling methods such as Markov Chain Monte Carlo (MCMC) or Langevin Dynamics (LD) can suffer from slow mixing times there is a growing interest in using normalizing flows in order to learn the transformation of a simple prior distribution to the given target distribution. Here we propose a generalized and combined approach to sample target densities: Stochastic Normalizing Flows (SNF) – an arbitrary sequence of deterministic invertible functions and stochastic sampling blocks. We show that stochasticity overcomes expressivity limitations of normalizing flows resulting from the invertibility constraint, whereas trainable transformations between sampling steps improve efficiency of pure MCMC/LD along the flow. By invoking ideas from non-equilibrium statistical mechanics we derive an efficient training procedure by which both the sampler's and the flow's parameters can be optimized end-to-end, and by which we can compute exact importance weights without having to marginalize out the randomness of the stochastic blocks.

Flow-based architectures have recently proved to be an efficient tool for numerical simulations of Effective String Theories regularized on the lattice that otherwise cannot be efficiently sampled by standard Monte Carlo methods. In this work we use Stochastic Normalizing Flows, a state-of-the-art deep-learning architecture based on non-equilibrium Monte Carlo simulations, to study different effective string models. After testing the reliability of this approach through a comparison with exact results for the Nambu-Got\={o} model, we discuss results on observables that are challenging to study analytically, such as the width of the string and the shape of the flux density. Furthermore, we perform a novel numerical study of Effective String Theories with terms beyond the Nambu-Got\={o} action, including a broader discussion on their significance for lattice gauge theories. These results establish the reliability and feasibility of flow-based samplers for Effective String Theories and pave the way for future applications on more complex models.

Deep generative models for approximating complicated and often high-dimensional probability distributions became a rapidly developing research field. Normalizing flows are a popular subclass of these generative models. They can be used to model a target distribution by a simpler latent distribution which is usually the standard normal distribution. In this paper, we are interested in finite normalizing flows which are basically concatenations of learned diffeomorphisms. The parameters of the diffeomorphism are adapted to the target distribution by minimizing a loss functions. To this end, the diffeomorphism must have a tractable Jacobian determinant. For the continuous counterpart of normalizing flows, we refer to the overview paper [43] and the references therein. Suitable architectures of finite normalizing flows include invertible residual neural networks (ResNets) [7, 11, 22], (coupling-based) invertible neural networks (INNs) [4, 14, 29, 34, 40] and autoregessive flows [13, 15, 26, 38].

Normalizing flows are popular generative learning methods that train an invertible function to transform a simple prior distribution into a complicated target distribution. Here we generalize the framework by introducing Stochastic Normalizing Flows (SNF) - an arbitrary sequence of deterministic invertible functions and stochastic processes such as Markov Chain Monte Carlo (MCMC) or Langevin Dynamics. This combination can be powerful as adding stochasticity to a flow helps overcoming expressiveness limitations of a chosen deterministic invertible function, while the trainable flow transformations can improve the sampling efficiency over pure MCMC. Key to our approach is that we can match a marginal target density without having to marginalize out the stochasticity of traversed paths. Invoking ideas from nonequilibrium statistical mechanics, we introduce a training method that only uses conditional path probabilities. We can turn an SNF into a Boltzmann Generator that samples asymptotically unbiased from a given target density by importance sampling of these paths. We illustrate the representational power, sampling efficiency and asymptotic correctness of SNFs on several benchmarks.