Diederik Kingma
TL;DR Diederik P. Kingma is a Dutch computer scientist, creator of the Variational Autoencoder, co-author of the Adam optimizer, and a co-founder of OpenAI, whose work reshaped modern generative AI.
Diederik Kingma by Sora
Diederik P. Kingma is a pioneering machine learning researcher whose ideas underpin today’s generative AI. Best known as the originator of the Variational Autoencoder (VAE) and co-author of the Adam optimization algorithm, Kingma helped make deep generative modeling practical and scalable. Born in the Netherlands, he earned a Ph.D. in machine learning at the University of Amsterdam under Max Welling. His 2013 preprint and 2014 paper on VAEs created a bridge between probabilistic modeling and deep neural networks, enabling models that learn latent structure and synthesize high-fidelity data.
Kingma’s work spans academia and industry. He has conducted research at Google Brain on large-scale optimization and generative models and at OpenAI, where he contributed to advancing foundation models. As a co-founder of OpenAI, he was part of the original group that set the lab’s mission and early technical direction focused on beneficial AGI. His research influence extends into diffusion models, representation learning, and the training recipes used across state-of-the-art systems.
Co-founder of OpenAI, helping define its early mission and research program
Created the Variational Autoencoder (VAE), a cornerstone of modern generative modeling
Co-authored the Adam optimizer, a default training method for deep neural networks
Ph.D. in Machine Learning, University of Amsterdam, supervised by Max Welling
Research at Google Brain and OpenAI, advancing scalable generative and optimization methods
Highly cited publications, shaping representation learning and probabilistic deep learning
Influence on diffusion and latent variable models, informing today’s text, image, and audio generators
