ART-AI and Centre for AI Seminar
We are delighted to welcome Leena Chennuru Vankadara, from University College London (UCL), for a joint ART-AI and Centre for AI seminar entitled ‘Towards a theory of scaling in deep learning.’ Abstract and Bio below.
Date: Wednesday 9 September 2026
Time: 15:15pm to 16:05pm (GMT)
Location: University of Bath 1W 3.103 Online via Microsoft Teams
For more information or if you would like to attend, please e-mail [email protected].
Title
Towards a theory of scaling in deep learning
Abstract
This talk explores scaling theory as a principled framework for understanding the role of scale in modern deep learning. While increases in model size, data, and computational resources often lead to predictable improvements in performance, they can also move models into qualitatively different operating regimes. Dr Leena Vankadara will discuss how performance gains depend not only on scale itself, but also on the ways in which models and training procedures are scaled.
Focusing on scaling limits as a natural lens for large-scale learning, she will show how these limits can be used to derive principled scaling rules while providing insights into the empirical behaviour of practical, finite-width neural networks. The talk highlights how scaling theory can help bridge the gap between theoretical understanding and the observed successes of modern deep learning systems.
Bio
Dr Leena Chennuru Vankadara is a Lecturer at the Gatsby Computational Neuroscience Unit at University College London (UCL). Prior to joining UCL, she was an Applied Scientist in the AGI Foundations Lab at Amazon, where she conducted fundamental research into the theory and science of scaling large language models.
She completed her PhD at the International Max Planck Research School for Intelligent Systems, working with Debarghya Ghoshdastidar at the Technical University of Munich and Ulrike von Luxburg at the University of Tübingen. Her doctoral thesis was recognised with the Wilhelm Schickard Dissertation Award for Outstanding Dissertation. During her PhD, she also spent six months at Amazon Research’s Causality Lab.
Dr Vankadara’s research focuses on developing efficient, reliable and trustworthy machine learning models through a deeper understanding of the theoretical foundations of deep learning and causal learning. Her work seeks to address fundamental questions in machine learning by developing a unified theory of scaling, investigating learning dynamics and generalisation in deep learning, and advancing formal frameworks for causal reasoning and the evaluation of causal models. Through this research, she aims to provide new theoretical foundations that can explain the behaviour of increasingly powerful AI systems.
