Scientific Organisers
-
Neill Campbell, University of Bath
-
James Foster (Lead organiser), University of Bath
-
Tony Shardlow, University of Bath
-
Kartic Subr, University of Edinburgh
-
Yue Wu, University of Strathclyde
About:
In recent years, the field of machine learning (ML) has seen tremendous progress, with many breakthroughs directly connected to the well-studied mathematical theory of Stochastic Differential Equations (SDEs). This increasingly fruitful relationship between SDEs and ML has produced several state-of-the-art innovations, ranging from Langevin algorithms in Bayesian learning to score-based diffusion models in computer vision.
This workshop aimed to bring the SDE and ML communities closer together and “sow the seeds” for future interdisciplinary and impactful research. The following general themes were explored:
-
SDE-inspired learning algorithms and architectures
-
Computational or learning-based algorithms for SDEs
-
Theoretical connections between SDEs and machine learning
-
Applications and areas of opportunity between disciplines
Resources
Programme
| Monday 3 June 2024 | ||
| Registration and Refreshments | ||
| Welcome and Housekeeping | ||
| Mini-course 1 - Desmond Higham, University of Edinburgh | Introduction to the Numerical Simulation of SDEs | |
| Lunch | ||
| Mini-course 2 - Fabio De Sousa Ribeiro, Imperial College London | Demystifying Diffusion Models | |
| Break and Discussion | ||
| Mini-course 3 - Andraž Jelinčič, University of Bath | Using Diffrax for efficient GPU-accelerated SDE simulation | |
| Tuesday 4 June 2024 | ||
| Konstantinos Zygalakis, University of Edinburgh | Talk Title TBC | |
| Break and Discussion | ||
| Tiffany Vlaar, University of Glasgow | Constrained and Partitioned Training of Neural Networks | |
| Break and Discussion | ||
| Robert Gruhlke, FU Berlin | Generative modelling with Tensor Train approximations of Hamilton–Jacobi–Bellman equations | |
| Lunch | ||
| Alexander Lobbe, Imperial College London | Generative Modelling of Stochastic Parametrisations for Geophysical Fluid Dynamics | |
| Break and Discussion | ||
| Yating Liu, Université Paris-Dauphine | Application of the optimal quantization and K-means clustering to the simulation of the McKean-Vlasov equation | |
| Break and Discussion | ||
| Teo Deveney, University of Bath | Closing the ODE-SDE gap in score-based diffusion models through the Fokker-Planck equation | |
| Break and Discussion | ||
| Terry Lyons, University of Oxford | Talk Title TBC | |
| Break and Discussion | ||
| Welcome Reception & Poster Session, hosted at ICMS | ||
| Public Lecture, hosted in G.03 (ground floor), Terry Lyons, University of Oxford | Signatures of Streams | |
| Wednesday 5 June 2024 | ||
| Desmond Higham, University of Edinburgh | Stability Issues for Diffusion Models in Generative AI | |
| Break and Discussion | ||
| Georgios Batzolis, University of Cambridge | Variational Diffusion Auto-encoder: Latent Space Extraction from Pre-trained Diffusion Models | |
| Break and Discussion | ||
| Thomas Gaskin, University of Cambridge | Neural parameter calibration for large-scale systems | |
| Lunch | ||
| Mini-course 4 - Grigoris Pavliotis, Imperial College London | Langevin-based sampling schemes | |
| Free afternoon , (Guided walk around the city) | ||
| Thursday 6 June 2024 | ||
| Neil Chada, Heriot-Watt University | Unbiased Kinetic Langevin Monte Carlo | |
| Break and Discussion | ||
| Benedict Leimkuhler, University of Edinburgh | Langevin and Adaptive Langevin Algorithms for Sampling and Optimisation in Machine Learning | |
| Break and Discussion | ||
| Lionel Riou-Durand, National Institute of Applied Sciences of Rouen | Metropolis Adjusted Langevin Trajectories: a robust alternative to Hamiltonian Monte Carlo | |
| Lunch | ||
| Josh Williams, STFC Hartree Centre | Modelling particle-laden turbulent flows with neural stochastic differential equations | |
| Break and Discussion | ||
| Irene Tubikanec, Johannes Kepler University Linz | Network inference in a stochastic multi-population neural mass model via approximate Bayesian computation | |
| Break and Discussion | ||
| Hao Ni, University College London | High Rank Path Development: an approach of learning the filtration of stochastic processes | |
| Break and Discussion | ||
| Workshop Dinner, hosted at Blonde Restaurant, 75 St. Leonard’s Street, Edinburgh EH8 7QR | ||
| Friday 7 June 2024 | ||
| Teresa Klatzer, University of Edinburgh | Bayesian Computation with Plug and Play Priors for Poisson Inverse Problems | |
| Break and Discussion | ||
| Mateusz Majka, Heriot-Watt University | Sampling, optimization, SDEs and gradient flows | |
| Break and Discussion | ||
| Grigoris Pavliotis, Imperial College London | Learning mean field models from data | |
| Lunch and End of Workshop |
