Igor Shilov
PHD Student at Imperial College London
London, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolIgor Shilov is a privacy-focused ML engineer and PhD student at Imperial College London with 11 years of industry experience building production-ready machine learning systems. He led development of Opacus at Meta (Facebook), a widely used PyTorch library for differential privacy, and has applied DP and federated learning across product and research settings. His work spans research, open-source engineering, and applied safety projects—most notably a privacy-preserving platform to combat non-consensual intimate image sharing—demonstrating skill moving algorithms into real-world systems. Igor contributes deep technical improvements to DP tooling (per-sample gradients, embeddings support, functorch bug fixes) and combines that with hands-on leadership of small engineering teams. Currently researching privacy-preserving ML in academia while holding an AI Safety fellowship, he blends rigorous research aims with practical, privacy-first engineering. Based in London, he is passionate about open source and building tools that make privacy-preserving ML accessible.
11 years of coding experience
12 years of employment as a software developer
Lomonosov Moscow State University
Russian, English
Github Skills (9)
pytorch10
machine-learning10
deep-learning10
differential-privacy10
testing10
model-validation9
rnn-model8
n8
nlp7
Programming languages (3)
Github contributions (5)
pytorch/opacus
Apr 2020 - Jan 2023
Training PyTorch models with differential privacy
Role in this project:
ML Engineer
ML EngineerContributions:5 releases, 208 reviews, 246 commits in 2 years 9 months
Contributions summary:Igor primarily contributes to the development and testing of differential privacy techniques within the PyTorch framework. Their work involves cleaning up and improving the accuracy of model validation exception messages, refactoring and testing per-sample gradient values, and adding support for affine transforms in layer normalization, group normalization. The user also contributed to enabling the use of embeddings, including example use cases for text classification and the fine-tuning of BERT. Finally, the user also addressed bugs related to gradient computation with functorch.
pytorchpytorch-modelsprivacydeep-learningprivacy-preserving-machine-learning
ffuuugor/moscowMrX
Aug 2015 - Jun 2017
Contributions:45 pushes, 1 branch in 1 year 11 months
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.