AI People Wiki

Builders & open source

Jonathan Frankle

Chief AI Scientist, Databricks

Jonathan Frankle is Chief AI Scientist at Databricks, where Frankle leads research projects on reinforcement learning, model training and agent evaluation; Frankle arrived through Databricks' acquisition of MosaicML, where Frankle was a member of the founding team. Frankle completed a PhD in computer science at MIT in 2023, and the main line of that research was the lottery ticket hypothesis, introduced in a 2018 paper with Michael Carbin, which studies how small neural networks can be during training. Frankle is a bylined author of Databricks' March 2024 post announcing the DBRX model and a co-author of the MosaicBERT paper on fast pretraining of a bidirectional encoder.12345

At a glance

Current role
Chief AI Scientist
Company
Databricks
Sector
Company
Education
MIT, Princeton
Work listed
3
Honors
1
Sources cited
8
Profile reviewed

Career

Jonathan Frankle completed a PhD in computer science at MIT in 2023, where the main line of research was the lottery ticket hypothesis. Frankle was a member of MosaicML's founding team and joined Databricks when it acquired MosaicML.12

As Chief AI Scientist at Databricks, Frankle leads research projects on reinforcement learning, model training, and agent evaluation. Frankle's homepage describes Mosaic Research at Databricks as sharing its work through blog posts, open-source repositories such as Composer, Streaming, LLM Foundry, and MegaBlocks, and open models such as DBRX and MPT.1

Frankle also works on technology policy, including with the OECD on implementing the AI Principles developed in 2019. Earlier, Frankle was the inaugural Staff Technologist at the Center on Privacy and Technology at Georgetown Law and co-developed a course on Computer Programming for Lawyers with Paul Ohm.1

Research

The lottery ticket hypothesis, introduced in a March 2018 paper with Michael Carbin, studies how large neural networks need to be in order to train. Frankle's homepage summarizes the finding that networks can be as small for much or all of training as they can be made after training. Follow-up work includes Linear Mode Connectivity and the Lottery Ticket Hypothesis (2019) and The Early Phase of Neural Network Training (2020).3178

More broadly, Frankle studies the behavior of practical neural networks empirically, posing hypotheses and testing them with large-scale experiments. At Databricks, Frankle is the first-listed author of the March 2024 post announcing DBRX, and co-authored the December 2023 MosaicBERT paper on a bidirectional encoder optimized for fast pretraining.145

Work and contributions

  • The Lottery Ticket Hypothesis Published research

    First author, with Michael Carbin, of "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (2018); Frankle's homepage describes this line of work as the main research of Frankle's PhD.31

  • DBRX

    First-listed of six bylined authors of Databricks' March 2024 post announcing DBRX, a mixture-of-experts language model built on the MegaBlocks research and open-source project.4

  • MosaicBERT Published research

    Co-author of "MosaicBERT: A Bidirectional Encoder Optimized for Fast Pretraining" (2023).5

Timeline

  1. Introduced the lottery ticket hypothesis in a paper with Michael Carbin.3
  2. Co-authored Linear Mode Connectivity and the Lottery Ticket Hypothesis.7
  3. Published The Early Phase of Neural Network Training with David Schwab and Ari Morcos.8
  4. Completed a PhD in computer science at MIT.1
  5. Co-authored MosaicBERT, a bidirectional encoder optimized for fast pretraining.5
  6. Co-authored Databricks' announcement of the DBRX open language model.4

Other documented roles

Roles the sources document that are not confirmed as current.

  • Member of the founding team, MosaicML12
  • PhD student in computer science (completed 2023), Massachusetts Institute of Technology1
  • Inaugural Staff Technologist, Center on Privacy and Technology, Georgetown Law1

Honors and recognition

  • ICLR Best Paper Award (The Lottery Ticket Hypothesis) · International Conference on Learning Representations (ICLR)2

Where they work

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

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

  1. Jonathan Frankle — homepageJonathan Frankle · Official page · accessed Oct 21, 2026
  2. Jonathan Frankle — Data + AI Summit 2026 speaker pageDatabricks · Official page · accessed Oct 21, 2026
  3. The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (arXiv 1803.03635)arXiv · Paper · published Mar 9, 2018 · accessed Oct 21, 2026
  4. Announcing DBRX: A new standard for efficient open source LLMsDatabricks · Announcement · published Mar 27, 2024 · accessed Oct 21, 2026
  5. MosaicBERT: A Bidirectional Encoder Optimized for Fast Pretraining (arXiv 2312.17482)arXiv · Paper · published Dec 29, 2023 · accessed Oct 21, 2026
  6. jfrankle (Jonathan Frankle) on GitHubGitHub · Official page · accessed Oct 21, 2026
  7. Linear Mode Connectivity and the Lottery Ticket HypothesisarXiv · Paper · published Dec 11, 2019 · accessed Oct 21, 2026
  8. The Early Phase of Neural Network TrainingarXiv · Paper · published Feb 24, 2020 · accessed Oct 21, 2026