Understanding memorisation of neural networks 

Understanding memorisation of neural networks

Cangxiong Chen, Trinity College Dublin.

8th of October 2026

Abstract: 

In the first half of the talk, I will present a brief overview of my past research in training-data leakage from neural networks and developing learning algorithms which protect against this leakage without sacrificing too much utility. In the second half, I will present my ongoing work with Alessio Benavoli on analysing memorisation of overparameterised neural networks using the theory of neural tangent kernels.