SeekamiDurk Kingma @dpkingma
Durk Kingma ranks #89 in the public Seekami AI Influence Top 100 and is classified as Researcher.
Data timing and version: ranking methodology
Why is this person notable?
Durk Kingma is a computer scientist working on scalable machine learning and generative models. Diederik P. Kingma, the name on many of his papers, is the same person. His research is best understood through contributions to model training and generative modeling rather than social visibility alone.
Career and project milestones
- 2013–2017 — Doctoral study at the University of Amsterdam, advised by Max Welling.
- 2015–2018 — Founding-team member and research scientist at OpenAI, working on algorithms.
- 2018–2024 — Research at Google Brain / DeepMind on generative models for text, images and video.
- 2024– — His website lists large-scale machine-learning research at Anthropic from 2024; this reflects the page reviewed for this update.
Durk Kingma · Personal biography ↗
VAE: learning with latent variables
Auto-Encoding Variational Bayes, submitted in 2013 with Max Welling, addresses inference when a generative model contains unobserved continuous variables. Approximate inference and reparameterization make the objective amenable to stochastic-gradient optimization. Start by distinguishing the generative model, inference model and objective: a VAE is more than an ordinary image-compression autoencoder.
Auto-Encoding Variational Bayes · Original paper ↗
Adam: an optimizer, not a model
Kingma and Jimmy Ba submitted Adam in 2014; it appeared at ICLR 2015. It uses estimates of first and second gradient moments to adapt parameter updates. It concerns optimization, not network architecture or the content a model generates. Read the update rule to distinguish learning rate, gradient history and the numerical-stability term.
Adam: A Method for Stochastic Optimization ↗
Glow: image generation through invertible transformations
Glow uses invertible 1×1 convolutions in a generative model. Its project presentation includes latent-space interpolation and attribute manipulation, not just generated images. It provides a route into flow-based modeling: how transformations connect data and latent representations, how sampling works and which changes can be explored in that space.
Glow · Research explanation and demonstrations ↗
A practical reading path
Suggested order: Adam → VAE → Glow. Begin with parameter updates, then latent-variable inference, then an alternative generative-modeling approach. No verified long-form biographical interview was selected for this update; the reading path uses the author’s biography and original work instead.
Questions readers ask
Are Durk and Diederik Kingma the same person?
Yes. His website identifies Durk as his nickname and Diederik as his formal name. Searching Diederik P. Kingma helps locate publication records.
Who co-authored Adam and the VAE paper?
Adam was co-authored with Jimmy Ba; Auto-Encoding Variational Bayes with Max Welling. These are collaborative contributions.
Why is his Seekami rank #89 despite major research contributions?
The #89 position describes account influence in the public sample. It is not a ranking of scientific contributions or paper quality. Social-network signals and the value of a research method measure different things.
Related people and organizations
OpenAI · Anthropic · Ian Goodfellow
These are editorial reading guides to original material. Related links do not imply endorsement.
Historical Seekami ranking data
Public research summary
| Field | Result |
|---|---|
| Rank band | Top 100 |
| Public category | Researcher |
| Useful signal | Research and model direction |
Peer position and reproducible findings
| Comparison | Result |
|---|---|
| Position in the public Top 100 | #89 of 100 |
| Researcher category position | #30 of 34 |
| Composite vs category average | -3.41 points (category average 89.05) |
| Strongest signal this release | Authority · 96.76 |
These findings describe relative position within the published snapshot. They support screening and verification, not inference about private attributes or future campaign results.
How to interpret the scores
The composite combines normalized public signals. Authority describes position in the public AI relationship network; public reach reflects observable scale; recent engagement reflects available activity signals in this release. Scores support relative comparison and are not ratings of truth, personal ability or campaign performance.
Creators nearby in the ranking
Source and limitations
This page uses the formal Seekami ranking snapshot. Profiles and public signals change, and rank is not a third-party endorsement. Read the methodology and data boundaries.