Ankit Siva

Hi, I'm Ankit, a software engineer at AWS. I work on privacy, fairness and explainability in algorithms.

Previously

• Software engineer on the Amazon SageMaker Debugger Team. Worked on future profiler capabilities.
• Software engineer on the Amazon SageMaker Clarify team. Worked on communicating bias and explainability results to users and operational excellence.
• Software engineering intern at Amazon Book Clubs. Worked on Book Clubs features.
• Undergraduate research advised by Rachel Cummings. Worked on breaking implementations of differential privacy and on private synthetic data generation.
• Research intern at Esri's Applications Prototype Lab. Investigated and implemented differential privacy mechanisms for geospatial data.
• Research intern at the MELT Lab, University of Minnesota. Designed the syntax, semantics and features of a tensor and linear algebra language extension of a custom port of C.

Publications

1. Private Synthetic Data for Multitask Learning and Marginal Queries. Joint work with Giuseppe Vietri, Cédric Archambeau,Sergul Aydore, William Brown, Michael Kearns, Aaron Roth, Shuai Tang and Steven Wu. In the proceedings of NeurIPS 2022.
2. Differentially Private Query Release Through Adaptive Projection. Joint work with Sergul Aydore, William Brown, Michael Kearns, Krishnaram Kenthapadi, Luca Melis and Aaron Roth. In the proceedings of ICML 2021. Source code available here.
3. Amazon SageMaker Clarify: Machine Learning Bias Detection and Explainability in the Cloud. Joint work with Michaela Hardt and the Amazon SageMaker Clarify team. In the proceedings of KDD 2021.
4. Differentially Private Synthetic Mixed-Type Data Generation For Unsupervised Learning. Joint work with Uthaipon Tantipongpipat, Chris Waites, Digvijay Boob and Rachel Cummings. In the proceedings of IISA 2021. Source code available here.
5. CHORUS is Porous: Attacking Implementations of Differential Privacy. My thesis, advised by Rachel Cummings.
6. 1 patent granted, 3 patent applications filed.

Links

• GitHub
• Google Scholar
• ArXiv
• ORCID

Get in touch

Email, Twitter, LinkedIn