Diederik P. Kingma

Career

Kingma worked as a junior research scientist in Yann LeCun's lab at New York University in 2009 and 2012, and from 2010 to 2012 was co-founder and technical lead of the company Advanza. From 2013 to 2017 Kingma did a PhD at the University of Amsterdam advised by Max Welling, on deep learning and generative models, with the thesis "Variational Inference and Deep Learning: A New Synthesis", and spent the summers of 2014 and 2015 collaborating at DeepMind.¹

Kingma was part of OpenAI's founding team in 2015 and led its Algorithms team, which focused on basic research, until 2018. From 2018 to 2024 Kingma was a research scientist at Google Brain and then Google DeepMind, heading projects on generative models for text, images and video, and since 2024 Kingma has worked on large-scale machine learning research at Anthropic.¹

Research

Several of Kingma's papers date from the doctoral years. "Auto-Encoding Variational Bayes" (2013), with Max Welling, introduced the Variational Autoencoder and its reparameterization approach to training latent-variable models, and a 2014 paper applied deep generative models to semi-supervised learning. "Adam: A Method for Stochastic Optimization" (2014), with Jimmy Ba, introduced an optimizer based on adaptive estimates of lower-order moments of the gradient.²⁸³

Later work developed flow-based and diffusion models. With Tim Salimans, Kingma wrote the weight normalization paper, and as first author introduced inverse autoregressive flow (2016) and Glow (2018), an invertible generative flow model. Kingma co-authored the score-based stochastic differential equation framework for generative modeling (2020), and as first author wrote "Variational Diffusion Models" (2021) and a 2023 paper relating common diffusion objectives to a weighted integral of evidence lower bounds.^910611712

Work and contributions

• Variational Autoencoder (Auto-Encoding Variational Bayes) Published research

  First author, with Max Welling, of the 2013 paper that introduced the Variational Autoencoder; winner of the inaugural ICLR Test of Time award in 2024.²⁴¹

• Adam optimizer Published research

  First author, with Jimmy Ba, of "Adam: A Method for Stochastic Optimization" (2014); winner of the ICLR 2025 Test of Time award.³⁵¹

• Glow and Variational Diffusion Models Published research

  Co-first author with Prafulla Dhariwal of "Glow: Generative Flow with Invertible 1x1 Convolutions" (2018) and first author, with Tim Salimans, Ben Poole, and Jonathan Ho, of "Variational Diffusion Models" (2021).⁶⁷¹

Timeline

1. Dec 2013Posted "Auto-Encoding Variational Bayes" with Max Welling, introducing the Variational Autoencoder.²
2. Dec 2014Posted "Adam: A Method for Stochastic Optimization" with Jimmy Ba.³
3. 2015Joined the founding team of OpenAI as a research scientist and led its Algorithms team.¹
4. Jun 2016Posted "Improving Variational Inference with Inverse Autoregressive Flow" as first author.¹⁰
5. 2017Completed a PhD cum laude at the University of Amsterdam, advised by Max Welling.¹
6. 2018Joined Google Brain as a research scientist.¹
7. Jul 2018Posted Glow, a generative flow with invertible 1x1 convolutions, with Prafulla Dhariwal.⁶
8. Nov 2020Co-authored "Score-Based Generative Modeling through Stochastic Differential Equations".¹¹
9. Jul 2021Posted "Variational Diffusion Models" as first author.⁷
10. Mar 2023Posted "Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation" with Ruiqi Gao.¹²
11. 2024Began research on large-scale machine learning at Anthropic.¹

Other documented roles

Roles the sources document that are not confirmed as current.

• Research Scientist (2018–2024), Google Brain / Google DeepMind¹
• Founding team member and Research Scientist (2015–2018); led the Algorithms team, OpenAI¹
• Junior Research Scientist (2009 and 2012), New York University (Yann LeCun's lab)¹
• Co-founder and technical lead (2010–2012), Advanza¹

Honors and recognition

• ICLR Test of Time Award (Adam) · International Conference on Learning Representations, 2025⁵¹
• ICLR Test of Time Award (Auto-Encoding Variational Bayes) · International Conference on Learning Representations, 2024⁴¹
• Dutch Data Science Award · Royal Holland Society of Sciences and Humanities, 2019¹
• ELLIS PhD Award · ELLIS (European Laboratory for Learning and Intelligent Systems), 2019¹
• Google European Doctoral Fellowship in Deep Learning · Google, 2015¹

Where they work

• Anthropic 34 people in the wiki

Related people

• Jascha Sohl-DicksteinAlso at Anthropic
• Amanda AskellAlso at Anthropic
• Nikhil ThoratAlso at Anthropic
• Arthur ConmyAlso at Anthropic
• Boris ChernyAlso at Anthropic
• Stefano ErmonImage, video, and 3D generation · Machine-learning theory
• Yang SongImage, video, and 3D generation · Machine-learning theory
• Peiyuan ZhangImage, video, and 3D generation
• Albert GuMachine-learning theory
• Alex SmolaMachine-learning theory