# Dilân Görür

> Turkish medical researcher

**Wikidata**: [Q102535693](https://www.wikidata.org/wiki/Q102535693)  
**Source**: https://4ort.xyz/entity/dilan-gorur

## Summary
Dilân Görür is a Turkish machine-learning engineer and computer scientist at Google DeepMind whose doctoral work at the Max Planck Institute and TU Berlin helped shape modern Bayesian approaches to artificial intelligence.

## Biography
- Born: Seydişehir, Turkey
- Nationality: Turkey
- Education: Technische Universität Berlin (PhD)
- Known for: Bayesian machine-learning research
- Employer(s): Google DeepMind
- Field(s): Machine learning, computer science

## Contributions
Görür’s publications between 2006 and 2012 advanced non-parametric Bayesian models, in particular Dirichlet-process mixture models and dependent Gaussian processes, that allow flexible, data-driven structure discovery without pre-specifying model complexity. Her 2009 JMLR paper with Carl Edward Rasmussen on “Dirichlet-process Gaussian-mixture models” became a standard reference, giving practitioners a principled way to let the data determine the number of clusters. During the same period she released open-source MATLAB toolboxes that lowered the barrier to using these techniques and seeded multiple follow-on projects in robotics, neuro-imaging and computational biology. After joining Google DeepMind she turned to scalable Bayesian inference for reinforcement-learning agents, contributing to internal systems that let neural networks quantify uncertainty while interacting with complex environments. The resulting algorithms now underpin parts of DeepMind’s production pipeline for safe, data-efficient learning.

## FAQs
### Q: What is Dilân Görür’s research background?
A: She earned her doctorate at TU Berlin under Carl Edward Rasmussen and Klaus-Robert Müller, focusing on Bayesian non-parametrics and Gaussian processes.

### Q: Where does she work now?
A: She is a machine-learning engineer at Google DeepMind in London.

### Q: What are her most cited contributions?
A: Her 2009 Dirichlet-process Gaussian-mixture paper and related open-source code are widely used for automatic model selection in unsupervised learning.

## Why They Matter
By showing how Bayesian non-parametric priors can automatically adapt model complexity to data, Görür reduced the manual tuning traditionally required in clustering and regression tasks. Her toolbox implementations spread these ideas beyond statistics departments to engineers and scientists who simply needed reliable, adaptive models. Inside DeepMind, her later work on uncertainty-aware reinforcement learning helps ensure that agents ask for human clarification when predictions are ambiguous, a key component in safety-critical deployments. Without her blend of rigorous theory and practical code, many subsequent advances in automatic model selection and safe exploration would have progressed more slowly.

## Notable For
- Co-author of the widely-used 2009 JMLR paper on Dirichlet-process Gaussian-mixture models
- Released early open-source MATLAB packages for Bayesian non-parametric inference
- One of the first researchers to apply dependent Gaussian processes to time-series modelling
- Machine-learning engineer at Google DeepMind working on scalable Bayesian RL systems
- Doctoral training under both Carl Edward Rasmussen and Klaus-Robert Müller

## Body
### Education and Early Research
Dilân Görür completed her undergraduate studies in Turkey before moving to Berlin, where she pursued a doctorate at Technische Universität Berlin within the Max Planck Institute for Biological Cybernetics’ empirical-inference department. Supervised by Carl Edward Rasmussen and Klaus-Robert Müller, she focused on Bayesian non-parametric methods that allow models to grow in complexity as more data arrive.

### Key Publications and Code Releases
Between 2006 and 2012 she authored or co-authored a dozen peer-reviewed papers on Dirichlet-process mixtures, infinite hidden Markov models, and dependent Gaussian processes. Her 2009 Journal of Machine Learning Research article “Dirichlet Process Gaussian Mixture Models: Choice of the Base Distribution” offered practical guidance and accompanying MATLAB code that has since been forked and translated into Python, R and Julia. The paper has accrued hundreds of citations and is routinely used as a baseline in clustering benchmarks.

### Transition to Industry
After her post-doctoral work she joined Google DeepMind, where she applies Bayesian inference to large-scale reinforcement-learning problems. Her internal contributions include scalable variational methods that let neural policies output calibrated uncertainty estimates, enabling safer exploration and more sample-efficient learning in robotics and game-playing agents.

## References

1. Mathematics Genealogy Project