Max Welling

Some people arrive at machine learning through mathematics, others through engineering. Max Welling arrived through quantum physics. After completing his doctorate at Utrecht under the supervision of Gerard 't Hooft (Nobel Prize in Physics), this Dutch scientist discovered that the same principles governing subatomic particles could be applied to teach a computer to learn. That unusual blend of statistical physics and neural networks became his personal trademark, and it is what sets him apart from almost any other researcher of his generation.

The moment that changed everything came in 2013, when he and Diederik P. Kingma published "Auto-Encoding Variational Bayes" and co-invented Variational Autoencoders (VAEs). In simple terms: a VAE learns not just to compress data, but to understand the space of possibilities that generates it, and can then create new, coherent data from that space. That idea is the seed from which all the generative AI we use today grew, from image generators to language models. Before, neural networks learned to recognize; afterward, they learned to create. Welling was one of the architects of that shift.

His career bridges academia and industry with a naturalness few achieve. He directs AMLab (Amsterdam Machine Learning Lab) at the University of Amsterdam and is a Distinguished Scientist at Microsoft Research AI4Science, also based in Amsterdam. In 2017, Qualcomm acquired Scyfer BV, the startup he co-founded, and named him VP of Technology for their Netherlands operations. He has received the ECCV Koenderink Prize in 2010 and the ICML Test of Time Award in 2021 for work whose relevance only grew with the years. He is also a pioneer in Graph Neural Networks and group-equivariant networks, which allow AI to leverage symmetries of the physical world to learn from less data.

What strikes me most about Welling is that he never stopped asking a physicist's questions, even while working in AI. That underlying curiosity, of wanting to understand the "why" before the "how," is what produces ideas that last for decades. Every time you generate an image with AI or fine-tune a generative model, there is an invisible trace of that mindset working underneath. The lesson is not that you need to study quantum physics, but that the best ideas tend to be born when you cross borders that others do not cross.