In the last few years, large language models (LLMs) have undergone significant development: in their capabilities, in the variety of their applications, and in their prevalence in our world. They have evolved from relatively fluent text generators to proficient reasoning models with which most people interact on a daily basis.
Mathematical Institute, Monday 7 to Wednesday 9 September.
This workshop will bring together graduate students and researchers in algebraic geometry. On Tuesday, 8 September, we will have a joint session with the CALF seminar (aimed at UK PhD students in algebraic geometry), featuring contributed talks by PhD students from both the UK and abroad. There will also be talks by postdocs and more senior researchers.
Substantial under-detection of human rabies cases in the United States.
Kim, Y
Donnelly, C
Communications health
volume 1
issue 1
23
(29 Aug 2026)
doi:10.1038/s44528-026-00031-4
Robust Bayes-Assisted Conformal Prediction
Ashouritaklimi, K
Cortinovis, S
Caron, F
Louis Thomas Hardin, better known as Moondog, was a blind avant-garde American musician who worked in a variety of genres including jazz and classical, invented a series of his own instruments, influenced a bunch of other composers including the American minimalists, made a lot of his living by begging and briefly appeared in a cloak and horned helmet (he loved Norse mythology) on Sixth Avenue in New York.
But never mind that.
Climate-driven host ecology as a framework for zoonotic disease risk: Understanding transferability across arenavirus systems
Milne, G
Attfield, L
Mariën, J
Kirkpatrick, L
Leirs, H
Jones, K
Donnelly, C
Redding, D
Proceedings of the National Academy of Sciences of the United States of America
volume 123
issue 39
e2625913123
(29 Sep 2026)
doi:10.1073/pnas.2625913123
Initial observations in X-point target divertor discharges on MAST-U
Lonigro, N
Verhaegh, K
Harrison, J
Lipschultz, B
Bowman, C
Federici, F
Flanagan, J
Greenhouse, D
Moulton, D
Ryan, P
Scannell, R
Silburn, S
Wijkamp, T
Brida, D
Theiler, C
Team, T
Nuclear Fusion
volume 66
issue 10
106007
(01 Oct 2026)
doi:10.1088/1741-4326/ae948e
Mon, 09 Nov 2026
15:30 -
16:30
L3
One-Step Generative Modeling via Wasserstein Gradient Flows
RenYuan XU
(Stanford University)
Abstract
Diffusion models and flow-based methods have achieved strong results in image generation, but often rely on costly iterative sampling. We introduce W-Flow, a framework that compresses a Wasserstein gradient flow into a one-step neural generator. By minimizing the Sinkhorn divergence, W-Flow transports a reference distribution toward the data distribution through efficient optimal-transport updates that capture global distributional discrepancies. We prove that, under suitable assumptions, the finite-sample training dynamics converge to the continuous-time distributional dynamics. Empirically, W-Flow sets a new state of the art in one-step ImageNet generation, with improved mode coverage and domain transfer. We will also discuss extensions using alternative energy functionals and applications to post-training.