creators

Durk Kingma - Anthropic, VAE and Adam optimizer

Durk Kingma researches large-scale machine learning at Anthropic. He co-created the variational autoencoder and Adam, both ICLR Test of Time winners.

Diederik P. "Durk" Kingma is a Dutch machine learning researcher known for deep generative models and optimisation. His own site, his Scholar profile and his X bio, all read today, put him at Anthropic, where he has worked on large-scale machine learning since October 2024. With Max Welling he introduced the variational autoencoder in "Auto-Encoding Variational Bayes" (2013); with Jimmy Ba he introduced Adam, a default optimiser for neural networks (2014). Both papers won ICLR Test of Time awards, the VAE in 2024 and Adam in 2025. He took his PhD cum laude at the University of Amsterdam in 2017, advised by Welling, was part of OpenAI's founding team from 2015 to 2018, where he led the algorithms team, and was a research scientist at Google Brain and DeepMind from 2018 to 2024, leading work on generative models for text, image and video.

Specialization

generative models, variational inference, optimization, deep learning

Country

Netherlands