Mathematician Charles Fefferman on His Personal Journey with Machine Learning

In a recent talk at the Barcelona Mathematics and Machine Learning (b=M2L) Colloquium, renowned mathematician Charles Fefferman shared his personal encounters with machine learning, detailing how he arrived at two distinct results connected to the field. The first is a uniqueness theorem: under generic conditions, neural networks that produce identical outputs must share the same architecture and parameters. The second result tackles the problem of fitting a smooth (C^m) function to data. Fefferman's lecture provides insights into the mathematical foundations behind modern AI and highlights the deep interplay between traditional mathematics and machine learning.