Utku Evci
Research Scientist at Google DeepMind
Montreal, Quebec, Canada
Summary
👤
Senior🎓
Top SchoolUtku Evci is a research scientist at Google DeepMind with over a decade of experience building and evaluating machine learning systems. His work spans reinforcement learning, computer vision, and meta-learning, and he is an active open-source contributor to projects like Dopamine and Scenic. At Dopamine, he implemented new Keras-based Atari models, refactored the network backbone to tf.keras, and refined DQN and Rainbow implementations along with test improvements. In Scenic, he contributed model definitions and checkpointing enhancements for ResNet and ViT-MLP blocks, including checks for loading stability and activation tracing. His work on Meta-Dataset focused on debugging data pipelines, data augmentation, and episode sampling to strengthen few-shot learning. He earned an M.Sc. in Computer Science from NYU Courant and has completed multiple international exchanges, and is based in Montreal, Canada, bridging research and production-ready ML systems.
11 years of coding experience
4 years of employment as a software developer
Bachelor of Science (B.Sc.), Electrical and Electronics Engineering, Bachelor of Science (B.Sc.), Electrical and Electronics Engineering at Koç ÜniversitesiMaster of Science (M.Sc.), Computer Science, 3.92, Master of Science (M.Sc.), Computer Science, 3.92 at New York UniversityExchange Student, Computer Engineering, Exchange Student, Computer Engineering at Nanyang Technological UniversityExchange Student, Elektrik ve Elektronik Mühendisliği, Exchange Student, Elektrik ve Elektronik Mühendisliği at Technische Universität München
Turkish, İngilizce, German, İtalyanca
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (22)
transformers10
python10
machine-learning10
reinforcement-learning10
flax10
keras10
tensorflow210
deep-learning10
tensorflow10
resnet10
ai10
computer-vision10
jax10
data-augmentation10
deep-learning-ai10
Programming languages (5)
Github contributions (5)
google-research/meta-dataset
Apr 2019 - Feb 2022
A dataset of datasets for learning to learn from few examples
Role in this project:
ML Engineer
ML EngineerContributions:38 commits, 1 PR, 9 pushes in 2 years 10 months
Contributions summary:Utku's commits primarily focused on debugging and improving the data pipeline within the meta-dataset project, specifically addressing issues related to data augmentation and episode sampling. These changes involved modifications to the code related to gin configurations for episode descriptions, as well as the implementation of runtime error handling to ensure the proper functioning of the data sampling methods. The commits showcase a detailed understanding of the data processing and sampling logic within the context of meta-learning.
meta-learningdatasetmachine-learningbenchmarklearning-to-learn
google/dopamine
Aug 2019 - Sep 2022
Dopamine is a research framework for fast prototyping of reinforcement learning algorithms.
Role in this project:ML Engineer
ML EngineerContributions:8 commits, 1 comment in 3 years 2 months
Contributions summary:Utku primarily focused on implementing and updating machine learning models within the Dopamine framework, as evidenced by the addition of new Keras-based models for Atari environments. They also refactored the network backbone to use `tf.keras` and updated existing agent implementations like DQN and Rainbow. Furthermore, the user's contributions involved addressing initialization issues and test suite enhancements for Rainbow and Implicit Quantile Networks.
reinforcement-learning-algorithmsfast-prototypinggooglereinforcement-learningprototyping
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.