Publications

2025

1. Implicit Regularisation in Diffusion Models: An Algorithm-Dependent Generalisation Analysis

   Tyler Farghly, Patrick Rebeschini, George Deligiannidis, and Arnaud Doucet

   2025

   arXiv
2. NeurIPS

   Rao-Blackwellised Reparameterisation Gradients

   Kevin Lam, Thang Bui, George Deligiannidis, and Yee Whye Teh

   Accepted at NeurIPS 2025, 2025

   arXiv
3. NeurIPS

   Schrödinger Bridge Matching for Tree-Structured Costs and Entropic Wasserstein Barycentres

   Samuel Howard, Peter Potaptchik, and George Deligiannidis

   Accepted at NeurIPS 2025, 2025

   arXiv
4. NeurIPS

   Diffusion Models and the Manifold Hypothesis: Log-Domain Smoothing is Geometry Adaptive

   Tyler Farghly, Peter Potaptchik, Samuel Howard, George Deligiannidis, and Jakiw Pidstrigach

   Accepted at NeurIPS 2025, 2025
5. Implicit Regularisation in Diffusion Models: An Algorithm-Dependent Generalisation Analysis

   Tyler Farghly, Patrick Rebeschini, George Deligiannidis, and Arnaud Doucet

   arXiv preprint arXiv:2507.03756, 2025
6. Mixing Time Bounds for the Gibbs Sampler under Isoperimetry

   Alexander Goyal, George Deligiannidis, and Nikolas Kantas

   2025

   Abs arXiv

   We establish bounds on the conductance for the systematic-scan and random-scan Gibbs samplers when the target distribution satisfies a Poincare or log-Sobolev inequality and possesses sufficiently regular conditional distributions. These bounds lead to mixing time guarantees that extend beyond the log-concave setting, offering new insights into the convergence behavior of Gibbs sampling in broader regimes. Moreover, we demonstrate that our results remain valid for log-Lipschitz and log-smooth target distributions. Our approach relies on novel three-set isoperimetric inequalities and a sequential coupling argument for the Gibbs sampler.

7. ICML

   Conditioning Diffusions Using Malliavin Calculus

   Jakiw Pidstrigach, Elizabeth Baker, Carles Domingo-Enrich, George Deligiannidis, and Nikolas Nüsken

   In ICML 2025, 2025

   arXiv
8. COLT

   Linear Convergence of Diffusion Models Under the Manifold Hypothesis

   Peter Potaptchik, Iskander Azangulov, and George Deligiannidis

   In Proceedings of Thirty Eighth Conference on Learning Theory, 2025

   Abs PDF

   Score-matching generative models have proven successful at sampling from complex high-dimensional data distributions. In many applications, this distribution is believed to concentrate on a much lower d-dimensional manifold embedded into D-dimensional space; this is known as the manifold hypothesis. The current best-known convergence guarantees are either linear in D or polynomial (superlinear) in d. The latter exploits a novel integration scheme for the backward SDE. We take the best of both worlds and show that the number of steps diffusion models require in order to converge in Kullback-Leibler (KL) divergence is linear (up to logarithmic terms) in the intrinsic dimension d. Moreover, we show that this linear dependency is sharp.

9. Understanding the Generalization Error of Markov algorithms through Poissonization

   Benjamin Dupuis, Maxime Haddouche, George Deligiannidis, and Umut Simsekli

   2025

   HTML

2024

1. On importance sampling and independent Metropolis-Hastings with an unbounded weight function

   George Deligiannidis, Pierre E Jacob, El Mahdi Khribch, and Guanyang Wang

   2024

   arXiv
2. Differentiable Cost-Parameterized Monge Map Estimators

   Samuel Howard, George Deligiannidis, Patrick Rebeschini, and James Thornton

   2024

   arXiv
3. Convergence of Diffusion Models Under the Manifold Hypothesis in High-Dimensions

   Iskander Azangulov, George Deligiannidis, and Judith Rousseau

   2024

   arXiv
4. Ranking in Generalized Linear Bandits

   Amitis Shidani, George Deligiannidis, and Arnaud Doucet

   In Workshop on Recommendation Ecosystems: Modeling, Optimization and Incentive Design, 2024

   arXiv
5. JRSSb

   From denoising diffusions to denoising Markov models

   Joe Benton, Yuyang Shi, Valentin De Bortoli, George Deligiannidis, and Arnaud Doucet

   JRSS(b) (discussion paper), 2024

   HTML
6. Error Bounds for Flow Matching Methods

   Joe Benton, George Deligiannidis, and Arnaud Doucet

   Transactions on Machine Learning Research, 2024

   HTML
7. AISTATS

   On the expected size of conformal prediction sets

   Guneet S Dhillon, George Deligiannidis, and Tom Rainforth

   In AISTATS, 2024

   arXiv
8. Linear Convergence Bounds for Diffusion Models via Stochastic Localization

   Joe Benton, Valentin De Bortoli, Arnaud Doucet, and George Deligiannidis

   ICLR (spotlight), 2024

   HTML
9. ICML

   Particle Denoising Diffusion Sampler

   Angus Phillips, Hai-Dang Dau, Michael John Hutchinson, Valentin De Bortoli, George Deligiannidis, and 1 more author

   In ICML, 2024

   HTML
10. JMLR

    Uniform Generalization Bounds on Data-Dependent Hypothesis Sets via PAC-Bayesian Theory on Random Sets

    Benjamin Dupuis, Paul Viallard, George Deligiannidis, and Umut Simsekli

    JMLR, 2024

    HTML
11. AAP

    Quantitative uniform stability of the iterative proportional fitting procedure

    George Deligiannidis, Valentin Bortoli, and Arnaud Doucet

    Annals of Applied Probability, 2024

    HTML

2023

1. NeurIPS

   A Unified Framework for U-Net Design and Analysis

   Christopher Williams, Fabian Falck, George Deligiannidis, Chris C Holmes, Arnaud Doucet, and 1 more author

   NeurIPS, 2023

   HTML
2. Wide stochastic networks: Gaussian limit and PAC-Bayesian training

   Eugenio Clerico, George Deligiannidis, and Arnaud Doucet

   In International Conference on Algorithmic Learning Theory, 2023

   HTML
3. ICML

   Generalization Bounds with Data-dependent Fractal Dimensions

   Benjamin Dupuis, George Deligiannidis, and Umut Simsekli

   ICML, 2023

   HTML

2022

1. AISTATS

   Neural Score matching for high-dimensional causal inference

   Oscar Clivio, Fabian Falck, Lehmann Brieuc, George Deligiannidis, and Chris Holmes

   In AISTATS, 2022

   HTML
2. AISTATS

   Conditional Gaussian PAC-Bayes

   Eugenio Clerico, George Deligiannidis, and Arnaud Doucet

   In AISTATS, 2022

   HTML
3. COLT

   Chained Generalisation Bounds

   Eugenio Clerico, Amitis Shidani, George Deligiannidis, and Arnaud Doucet

   In COLT, 2022

   HTML
4. NeurIPS

   A Continuous Time Framework for Discrete Denoising Models

   Andrew Campbell, Joe Benton, Valentin De Bortoli, Tom Rainforth, George Deligiannidis, and 1 more author

   NeurIPS 2022 (oral), 2022

   HTML
5. NeurIPS

   A Multi-Resolution Framework for U-Nets with Applications to Hierarchical VAEs

   Fabian Falck, Christopher Williams, Dominic Danks, George Deligiannidis, Christopher Yau, and 3 more authors

   NeurIPS 2022 (oral), 2022

   HTML
6. Conditional Simulation Using Diffusion Schrödinger Bridges

   Yuyang Shi, Valentin De Bortoli, George Deligiannidis, and Arnaud Doucet

   In UAI 2022, 2022

   HTML
7. JRSSb

   Non-reversible parallel tempering: a scalable highly parallel MCMC scheme

   Saifuddin Syed, Alexandre Bouchard-Côté, George Deligiannidis, and Arnaud Doucet

   JRSS(B), 2022

   HTML

2021

1. AISTATS

   Stable resnet

   Soufiane Hayou, Eugenio Clerico, Bobby He, George Deligiannidis, Arnaud Doucet, and 1 more author

   In AISTATS (Oral), 2021

   HTML
2. ICML

   Differentiable particle filtering via entropy-regularized optimal transport

   Adrien Corenflos, James Thornton, George Deligiannidis, and Arnaud Doucet

   In ICML (Long Oral), 2021

   HTML
3. NeurIPS

   Fractal structure and generalization properties of stochastic optimization algorithms

   Alexander Camuto, George Deligiannidis, Murat A Erdogdu, Mert Gurbuzbalaban, Umut Simsekli, and 1 more author

   NeurIPS (Spotlight), 2021

   HTML
4. AAP

   Randomized Hamiltonian Monte Carlo as scaling limit of the bouncy particle sampler and dimension-free convergence rates

   George Deligiannidis, Daniel Paulin, Alexandre Bouchard-Côté, and Arnaud Doucet

   Annals of Applied Probability, 2021

   HTML
5. Boundary of the Range of a random walk and the Fölner property

   George Deligiannidis, Sebastien Gouëzel, and Zemer Kosloff

   Electronic Journal of Probability, 2021

   HTML
6. Random walk algorithm for the Dirichlet problem for parabolic integro-differential equation

   George Deligiannidis, S Maurer, and Michael V Tretyakov

   BIT Numerical Mathematics, 2021

2020

1. ICML

   Relaxing bijectivity constraints with continuously indexed normalising flows

   Rob Cornish, Anthony Caterini, George Deligiannidis, and Arnaud Doucet

   In ICML, 2020

   HTML
2. NeurIPS

   Hausdorff dimension, heavy tails, and generalization in neural networks

   Umut Simsekli, Ozan Sener, George Deligiannidis, and Murat A Erdogdu

   NeurIPS (Spotlight), 2020

   HTML
3. Ensemble rejection sampling

   George Deligiannidis, Arnaud Doucet, and Sylvain Rubenthaler

   2020

   arXiv
4. AOS

   Controlled sequential monte carlo

   Jeremy Heng, Adrian N Bishop, George Deligiannidis, and Arnaud Doucet

   Annals of Statistics, 2020

   HTML
5. Large sample asymptotics of the pseudo-marginal method

   Sebastian M Schmon, George Deligiannidis, Arnaud Doucet, and Michael K Pitt

   Biometrika, 2020

   HTML
6. Unbiased Markov chain Monte Carlo for intractable target distributions

   Lawrence Middleton, George Deligiannidis, Arnaud Doucet, and Pierre E Jacob

   Electronic Journal of Statistics, 2020

   HTML
7. Efficient irreversible Monte Carlo samplers

   Fahim Faizi, George Deligiannidis, and Edina Rosta

   Journal of Chemical Theory and Computation, 2020
8. Simulated tempering with irreversible Gibbs sampling techniques

   Fahim Faizi, Pedro J Buigues, George Deligiannidis, and Edina Rosta

   The Journal of Chemical Physics, 2020

2019

1. ICML

   Scalable Metropolis-Hastings for exact Bayesian inference with large datasets

   Robert Cornish, Paul Vanetti, Alexandre Bouchard-Côté, George Deligiannidis, and Arnaud Doucet

   In ICML (Long oral), 2019

   HTML
2. AISTATS

   Bernoulli race particle filters

   Sebastian M Schmon, Arnaud Doucet, and George Deligiannidis

   In AISTATS, 2019

   HTML
3. AISTATS

   Unbiased smoothing using particle independent Metropolis-Hastings

   Lawrece Middleton, George Deligiannidis, Arnaud Doucet, and Pierre E Jacob

   In AISTATS 2019, (Oral), 2019

   HTML
4. AOS

   Exponential Ergodicity of the Bouncy Particle Sampler

   George Deligiannidis, Alexandre Bouchard-Côté, and Arnaud Doucet

   Annals of Statistics, 2019

   HTML

2018

1. AAP

   Which ergodic averages have finite asymptotic variance?

   George Deligiannidis, and Anthony Lee

   Annals of Applied Probability, 2018

   HTML

2017

1. AOP

   Relative Complexity of Random Walks in Random Scenery in the absence of a weak invariance principle for the local times

   George Deligiannidis, and Zemer Kosloff

   Annals of Probability, 2017

   HTML

2016

1. Optimal bounds for the variance of self-intersection local times

   George Deligiannidis, and Sergey Utev

   International Journal of Stochastic Analysis, 2016

   HTML

2015

1. Efficient implementation of Markov chain Monte Carlo when using an unbiased likelihood estimator

   Arnaud Doucet, Michael K Pitt, George Deligiannidis, and Robert Kohn

   Biometrika, 2015

   HTML
2. JRSSb

   The Correlated Pseudo-Marginal Method

   George Deligiannidis, Arnaud Doucet, and Michael K Pitt

   JRSS(B), 2015

   HTML

2014

1. EJP

   Asymptotic variance of stationary reversible and normal Markov processes

   George Deligiannidis, Magda Peligrad, and Sergey Utev

   Electronic Journal of Probability, 2014

   HTML

2013

1. AOP

   Variance of partial sums of stationary sequences

   George Deligiannidis, and Sergey Utev

   Annals of Probability, 2013

   HTML

2011

1. Asymptotic variance of the self-intersections of stable random walks using Darboux-Wiener theory

   George Deligiannidis, and Sergei A. Utev

   Siberian mathematical journal, 2011

   HTML

2009

1. Optimal Stopping for processes with independent increments, and applications

   G Deligiannidis, H Le, and S Utev

   Journal of Applied Probability, 2009

   HTML