Ankur Taly
Senior Staff Research Scientist at Google DeepMind
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAnkur Taly is a Senior Staff Research Scientist with 13 years of experience building trustworthy, explainable, and data-centric AI systems at the intersection of research and product. Based in Mountain View, he has driven model and data quality efforts for Google Cloud and Gemini/Bard, led cross-team initiatives on grounding and factuality for LLMs, and taught Stanford’s CS329T course on Trustworthy Machine Learning. He co-developed widely adopted explainability methods (Integrated Gradients) and has a track record of translating theory into product features and Cloud services. His open-source contributions include improving the HotFlip counterfactual component in the popular LIT interpretability tool, reflecting a focus on practical tools for model analysis. Trained at Stanford (PhD) and IIT Bombay, he combines rigorous formal work in security and program analysis with applied ML solutions that improve production reliability. He often bridges evaluation, human-in-the-loop workflows, and automated raters to make AI systems more robust and auditable.
13 years of coding experience
18 years of employment as a software developer
Indian Institute of Technology BombayPhD Computer Science, PhD Computer Science at Stanford University
English, Hindi
Github Skills (9)
algorithm10
algorithms10
machine-learning10
interpretation10
python10
natural-language-processing10
tensorflow9
visualizations7
visualization7
Programming languages (6)
Github contributions (5)
PAIR-code/lit
Apr 2021 - Jul 2021
The Learning Interpretability Tool: Interactively analyze ML models to understand their behavior in an extensible and framework agnostic interface.
Role in this project:
ML Engineer
ML EngineerContributions:15 commits in 3 months
Contributions summary:Ankur primarily contributed to the development of the HotFlip component within the LIT (Learning Interpretability Tool) repository. Their work involved implementing and refining the HotFlip algorithm, a method for generating counterfactual examples to understand machine learning models. Key contributions include updates to the HotFlip algorithm, such as ensuring minimal token flips and supporting regression models. They also focused on improving the algorithm's efficiency and usability through code refactoring and adding new features.
ml-modelspythoninterpretabilitydata-sciencedeep-learning
ankurtaly/Integrated-Gradients
Nov 2016 - Feb 2022
Attributing predictions made by the Inception network using the Integrated Gradients method
Contributions:1 review, 30 commits, 3 PRs in 5 years 4 months
inception-networkgradientsintegratedintegrated-gradientsmethod
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.