Event Details

This three-day workshop focuses on the theoretical understanding of scaling laws in large neural network models, whose empirical principles have already been instrumental in modern AI breakthroughs.

Modern neural networks operate at an unprecedented scale across numerous dimensions: model size, data volume, amount of compute resources and many more. A central research problem is to understand how neural network performance scales with these dimensions, which in turn informs on how to train networks optimally at these large scales. In recent years, empirical heuristics of scaling have arguably been one of the main drivers behind the successes of Large Language Models (LLMs). Meanwhile, theoretical research on scaling laws has seen much fruitful progress, offering perspectives for empirical phenomena such as model collapse, emergence and training stability, while also providing concrete practical methods in areas such as hyperparameter tuning.

This workshop will bring together researchers working on the frontiers of theoretical scaling laws to discuss their insights about the field. The workshop will be the first of its kind in the UK, inspired by successes of similar workshops in the US and Europe. The first half of the workshop consists of introductory tutorials, with the aim of equipping attendees with basic tools for framing and understanding problems in this field. The second half will focus on latest research advances. The desired outcome is for UK researchers across academia and industry to learn about and participate in this active field of research, which has seen many fruitful empirical outcomes.

This workshop is jointly funded by Centre for Research in Statistical Methodology (CRiSM) at Warwick Statistics and the Prob_AI Hub.

Confirmed speakers:

• Blake Bordelon (Harvard)
• Leena C Vankadara (UCL)
• Lénaïc Chizat (EPFL)
• Mufan Li (Waterloo)
• Sam Smith (OpenAI)

Open until 31 March on a first-come-first-serve basis: https://warwick.ac.uk/fac/sci/statistics/news/probai-scaling-laws-2026/registration/