![[2602.20758] Deep unfolding of MCMC kernels: scalable, modular & explainable GANs for high-dimensional posterior sampling](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.20758] Deep unfolding of MCMC kernels: scalable, modular & explainable GANs for high-dimensional posterior sampling
Summary
This article presents a novel approach to integrating deep unfolding techniques with MCMC methods, enhancing the efficiency and interpretability of GANs for high-dimensional posterior sampling.
Why It Matters
The study addresses the computational challenges of traditional MCMC methods in Bayesian inference, offering a scalable and modular solution through GANs. This advancement could significantly improve the efficiency of posterior sampling in various applications, making it relevant for researchers and practitioners in machine learning and statistics.
Key Takeaways
- Introduces a new GAN architecture that enhances MCMC methods.
- Achieves high sampling accuracy while maintaining computational efficiency.
- Allows for modular design and interpretability in Bayesian computations.
- Demonstrates robustness to changes in likelihood parameters.
- Provides a supervised regularized Wasserstein GAN framework for training.
Computer Science > Machine Learning arXiv:2602.20758 (cs) [Submitted on 24 Feb 2026] Title:Deep unfolding of MCMC kernels: scalable, modular & explainable GANs for high-dimensional posterior sampling Authors:Jonathan Spence, Tobías I. Liaudat, Konstantinos Zygalakis, Marcelo Pereyra View a PDF of the paper titled Deep unfolding of MCMC kernels: scalable, modular & explainable GANs for high-dimensional posterior sampling, by Jonathan Spence and 2 other authors View PDF Abstract:Markov chain Monte Carlo (MCMC) methods are fundamental to Bayesian computation, but can be computationally intensive, especially in high-dimensional settings. Push-forward generative models, such as generative adversarial networks (GANs), variational auto-encoders and normalising flows offer a computationally efficient alternative for posterior sampling. However, push-forward models are opaque as they lack the modularity of Bayes Theorem, leading to poor generalisation with respect to changes in the likelihood function. In this work, we introduce a novel approach to GAN architecture design by applying deep unfolding to Langevin MCMC algorithms. This paradigm maps fixed-step iterative algorithms onto modular neural networks, yielding architectures that are both flexible and amenable to interpretation. Crucially, our design allows key model parameters to be specified at inference time, offering robustness to changes in the likelihood parameters. We train these unfolded samplers end-to-end using a supe...
