PeopleofSuperintelligence

Researcher

Harikrishna Narasimhan

Also known as Hari Narasimhan

Harikrishna Narasimhan is a machine-learning researcher at Google DeepMind whose work spans algorithmic fairness, constrained optimization and efficient language-model inference. He earned a computer science PhD at the Indian Institute of Science in 2015 with Shivani Agarwal and worked as a postdoctoral fellow at Harvard University with David C. Parkes from 2015 to 2018. His research moved from neural-network auction design and the TensorFlow Constrained Optimization library in 2019 and 2020 to language-model cascades and Faster Cascades via Speculative Decoding, which implements a cascade's deferral rule through speculative execution and received an ICLR 2025 Outstanding Paper Honorable Mention. As of October 2026, he is a Staff Research Scientist at Google DeepMind in Mountain View, focusing on inference efficiency in large language models.

Badges

Timeline

  1. 2012–2015

    A PhD at IISc

    From 2012 to 2015, he completes a PhD in computer science in the Department of Computer Science and Automation at the Indian Institute of Science in Bangalore with Shivani Agarwal, receiving a 2013 Google India PhD Fellowship in Machine Learning.

  2. 2015–2018

    A postdoc at Harvard

    From 2015 to 2018, Narasimhan works as a postdoctoral fellow at Harvard University with David C. Parkes.

  3. 2019

    Neural auctions and deterministic classifiers

    At ICML 2019, he coauthors Optimal Auctions through Deep Learning with David C. Parkes and colleagues, modeling an auction as a multi-layer neural network for the automated design of optimal auctions. At NeurIPS 2019, he coauthors On Making Stochastic Classifiers Deterministic, proving lower and upper bounds on how well a deterministic classifier can approximate a stochastic one.

  4. 2020

    Fairness and constrained classification

    In February 2020, with Andrew Cotter, Maya R. Gupta, Heinrich Jiang and other authors of the TensorFlow Constrained Optimization library, he coauthors a Google Research blog post on setting fairness goals. At NeurIPS 2020, his paper Consistent Plug-in Classifiers for Complex Objectives and Constraints presents a consistent algorithm for constrained classification with objectives such as F-measure and G-mean and constraints such as demographic parity fairness and coverage.

  5. 2024

    Language-model cascades and NeurIPS chair

    At ICLR 2024, he coauthors Language Model Cascades: Token-Level Uncertainty And Beyond, which initiates a systematic study of deferral rules for language-model cascades. In 2024, he also publishes a longer journal paper on complex loss functions and constraints in the Journal of Machine Learning Research and serves as Senior Area Chair for NeurIPS 2024.

  6. 2024–2025

    Speculative cascades and an ICLR award

    In May 2024, Narasimhan and his coauthors post Faster Cascades via Speculative Decoding, designing speculative cascading techniques that implement a cascade's deferral rule through speculative execution. The paper receives an ICLR 2025 Outstanding Paper Honorable Mention in April 2025. In September 2025, he and Aditya Menon publish a Google Research blog post describing how speculative cascades combine speculative decoding with standard cascades to improve language-model efficiency and computational costs.

  7. As of October 2026

    Staff Research Scientist

    He is a Staff Research Scientist at Google DeepMind in Mountain View and focuses on improving inference efficiency in large language models.

Career

  1. NowGoogle DeepMindStaff Research Scientist
  2. 2015–2018Harvard UniversityPostdoctoral Fellow

Education

  1. 2015Indian Institute of SciencePhD

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Harikrishna Narasimhan. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Harikrishna Narasimhan yet. Send us a free-licensed photo.

  1. Stochastic Methods for Complex Performance Measures: A Tale of Two Families microsoft.com. Checked 22 October 2026.
  2. Harikrishna Narasimhan (Hari) hari-research.github.io. Checked 22 October 2026.
  3. Optimal Auctions through Deep Learning proceedings.mlr.press. Checked 22 October 2026.
  4. On Making Stochastic Classifiers Deterministic papers.nips.cc. Checked 22 October 2026.
  5. Setting Fairness Goals with the TensorFlow Constrained Optimization Library research.google. Checked 22 October 2026.
  6. Consistent Plug-in Classifiers for Complex Objectives and Constraints research.google. Checked 22 October 2026.
  7. Language Model Cascades: Token-Level Uncertainty And Beyond research.google. Checked 22 October 2026.
  8. Faster Cascades via Speculative Decoding arxiv.org. Checked 22 October 2026.
  9. Announcing the Outstanding Paper Awards at ICLR 2025 blog.iclr.cc. Checked 22 October 2026.
  10. Speculative cascades — A hybrid approach for smarter, faster LLM inference research.google. Checked 22 October 2026.