AI People Wiki

Builders & open source

David D. Cox

VP, AI Foundations; IBM Director, MIT-IBM Computing Research Lab, IBM

  • VP, AI Foundations; IBM Director, MIT-IBM Computing Research Lab, IBM Research21

David Cox is VP, AI Foundations at IBM Research and the IBM director of the MIT-IBM Computing Research Lab, which IBM Research's profile lists as the MIT-IBM Watson AI Lab. IBM Research describes Cox as responsible for the development and training of IBM's large language models, and Cox is a co-author of the Granite Code and Granite Embedding papers, both of which release their models under the Apache 2.0 license; the Granite Code paper describes sizes from 3 billion parameters suited to on-device, memory-constrained use. Before IBM, Cox was the John L. Loeb Associate Professor of the Natural Sciences and of Engineering and Applied Sciences at Harvard University, where Cox's lab started the open-source Hyperopt hyperparameter optimization library.123456

At a glance

Current role
VP, AI Foundations; IBM Director, MIT-IBM Computing Research Lab
Company
IBM
Sector
Company
Education
Harvard
Work listed
3
Honors
3
Sources cited
6
Profile reviewed

Career

Before joining IBM, Cox was the John L. Loeb Associate Professor of the Natural Sciences and of Engineering and Applied Sciences at Harvard University, with appointments in Computer Science, the Department of Molecular and Cellular Biology and the Center for Brain Science. Cox was also a Faculty Associate at the Berkman Klein Center for Internet and Society at Harvard Law School, and while at Harvard co-founded several AI startups, including DeepHealth, which developed FDA-cleared medical imaging software.12

At IBM Research, Cox is VP, AI Foundations, leading the innovation agenda for foundation models and responsible for the development and training of IBM's large language models. Cox is also IBM's director of the joint lab with MIT, which brings IBM researchers together with MIT faculty on fundamental AI research.12

Research and open-source work

Cox's research spans neuroscience and AI. With James Bergstra and Daniel Yamins, Cox wrote an ICML 2013 paper on automating hyperparameter search for vision architectures, and Cox's Harvard lab started the open-source projects Hyperopt and Triton, the GPU compute framework now known as OpenAI Triton.156

At IBM, Cox is a co-author of papers on the Granite model family, including the 2024 Granite Code models, which are released under Apache 2.0, a Granite function-calling model presented at EMNLP 2024, and the 2025 Granite Embedding models for dense and sparse retrieval in English and multilingual settings. Other recent co-authored work listed by IBM Research includes Activated LoRA (NeurIPS 2025) and Self-MoE (ICLR 2025).341

Work and contributions

  • IBM Granite models Open weights

    Listed among the authors of the Granite Code models paper (2024), which releases code models from 3 to 34 billion parameters under an Apache 2.0 license, and of the Granite Embedding models report (2025), which releases English and multilingual retrieval embedding models under Apache 2.0. IBM Research's profile describes Cox as responsible for the development and training of IBM's large language models.341

  • Hyperopt Open-source code

    IBM Research's profile states that Cox's lab at Harvard started Hyperopt, a BSD-licensed Python library for hyperparameter optimization; Cox co-authored, with James Bergstra and Daniel Yamins, the ICML 2013 paper "Making a Science of Model Search: Hyperparameter Optimization in Hundreds of Dimensions for Vision Architectures."156

  • Triton Open-source code

    IBM Research's profile states that Cox's lab at Harvard started Triton, now OpenAI Triton, a framework for accelerated compute on GPUs.1

Timeline

  1. Co-authored, with James Bergstra and Daniel Yamins, the ICML 2013 paper on hyperparameter optimization for vision architectures.5
  2. Co-authored the Granite function-calling model paper presented at EMNLP 2024.1
  3. Co-authored the Granite Code models paper, released with Apache 2.0 code models.3
  4. Co-authored the Activated LoRA paper on fine-tuned LLM intrinsics, presented at NeurIPS 2025.1
  5. Co-authored the Granite Embedding models report on retrieval embedding models.4

Other documented roles

Roles the sources document that are not confirmed as current.

  • John L. Loeb Associate Professor of the Natural Sciences and of Engineering and Applied Sciences, Harvard University12
  • Faculty Associate, Berkman Klein Center for Internet and Society, Harvard Law School12

Honors and recognition

  • Richard and Susan Smith Foundation Award for Excellence in Biomedical Research · Richard and Susan Smith Family Foundation12
  • Google Faculty Research Award in Computer Science · Google12
  • Roslyn Abramson Award for Excellence in Undergraduate Teaching · Harvard University12

Where they work

  • IBM 29 people in the wiki

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

6 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. David D. Cox — IBM ResearchIBM Research · Official page · accessed Oct 21, 2026
  2. David Cox — MIT-IBM Computing Research Lab people pageMIT-IBM Computing Research Lab · Official page · accessed Oct 21, 2026
  3. Granite Code Models: A Family of Open Foundation Models for Code Intelligence (arXiv 2405.04324)arXiv · Paper · published May 7, 2024 · accessed Oct 21, 2026
  4. Granite Embedding Models (arXiv 2502.20204)arXiv · Paper · published Feb 27, 2025 · accessed Oct 21, 2026
  5. Making a Science of Model Search: Hyperparameter Optimization in Hundreds of Dimensions for Vision Architectures (ICML 2013)Proceedings of Machine Learning Research · Paper · published 2013 · accessed Oct 21, 2026
  6. hyperopt/hyperopt README and LICENSE.txt (BSD-style license)hyperopt (GitHub repository) · Repository · accessed Oct 21, 2026