Textbook of AI

1968–, Computer scientist

Max Welling is a Dutch computer scientist whose 2013 paper with Diederik Kingma Auto-Encoding Variational Bayes introduced the variational autoencoder, a probabilistic generative model that combines an encoder network producing parameters of an approximate posterior with a decoder network reconstructing data from latent samples. The reparameterisation trick the paper introduced, sampling z = μ + σ ⊙ ε, with ε ~ N(0, I) so the gradient flows through the sampling step, made backpropagation through stochastic latent variables practical.

Welling has had a long career in physics-flavoured machine learning, contributing to MCMC methods, graph neural networks, equivariant networks (for physics applications), and most recently AI for scientific discovery. He held chairs at Amsterdam and UC Irvine and has been a Vice President of Technologies at Qualcomm.