AI People Wiki

Builders & open source

David Bau

Assistant Professor of Computer Science, Northeastern University

  • Assistant Professor of Computer Science, Northeastern University (Khoury College of Computer Sciences)1
  • Principal Investigator and Director (2024–present), National Deep Inference Fabric (NDIF)4
  • Director (2013–present), Pencil Code4

David Bau is an Assistant Professor of Computer Science at Northeastern University's Khoury College, where Bau's lab studies the structure and interpretation of deep networks. The lab launched the National Deep Inference Fabric (NDIF) project and points researchers to the open-source nnsight library, which lets users read and edit a model's internals; Bau is also a co-author of the ROME paper on locating and editing factual associations in GPT models. Bau's earlier industry work includes heading the team that created Google Talk, contributing to Google Image Search, and working on the Trident team for Microsoft Internet Explorer 4.123

At a glance

Current roles
Assistant Professor of Computer Science +2 more
Sector
University
Education
Cornell, Harvard, MIT
Work listed
2
Honors
3
Sources cited
5
Profile reviewed

Career

Bau's CV lists roles at Microsoft from 1993 to 2000, including work on the .NET Common Language Runtime and on the Trident engine behind Internet Explorer 4 to 6, followed by heading the design of the Weblogic Workshop developer tool at Crossgain and BEA Systems. At Google from 2004 to 2015, Bau built the team that created Google Talk, worked on how Google Search handles queries about people, and led Google Image Search teams on freshness, a continuous-results redesign and image clustering.4

Bau then completed a PhD in EECS at MIT, with the thesis "Dissection of Deep Neural Networks" advised by Antonio Torralba, and was a postdoctoral fellow in Martin Wattenberg's lab at Harvard. Bau joined Northeastern's Khoury College as an assistant professor in 2022 and since 2024 has directed the National Deep Inference Fabric, a research infrastructure project supported by the National Science Foundation. Since 2013 Bau has also directed Pencil Code, an educational programming system that bridges block-based and text-based code.41

Research

Bau's early interpretability papers include Network Dissection (CVPR 2017), a way to quantify how interpretable the units of a vision network are, and GAN Dissection (ICLR 2019), which applies the same idea to generative adversarial networks; a 2020 PNAS paper on the role of individual units and an ECCV 2020 paper on rewriting a deep generative model extended this work.4

Later work turned to language models. The NeurIPS 2022 paper on locating and editing factual associations in GPT (ROME) and its ICLR 2023 follow-up on mass-editing memory in a transformer locate where facts are stored and change them directly. The lab's NNsight and NDIF paper (ICLR 2025) describes a programming library and a remote inference service that let researchers read and alter the internals of large open-weight models.3425

Work and contributions

  • NDIF and nnsight Open-source code

    Bau's lab page announces the launch of the National Deep Inference Fabric (NDIF) and points to nnsight for trying it; Bau is the last-listed author of the repository's citation for "NNsight and NDIF: Democratizing Access to Foundation Model Internals". The nnsight repository is released under the MIT license.125

  • Locating and Editing Factual Associations in GPT (ROME) Published research

    Co-author of the 2022 paper, which uses causal interventions to locate the computations that mediate factual recall in autoregressive transformers and edits feed-forward weights to update specific factual associations.3

Timeline

  1. Began work at Google on Google Talk, creating the team behind the realtime communications product.4
  2. Began heading Google Image Search teams as a staff software engineer.4
  3. Began directing Pencil Code, an educational programming system.4
  4. Started PhD research in Antonio Torralba's lab at MIT CSAIL.4
  5. Presented Network Dissection as an oral paper at CVPR 2017.4
  6. Completed an MIT PhD with the thesis "Dissection of Deep Neural Networks" and began a postdoc at Harvard.4
  7. Joined Northeastern University's Khoury College as an Assistant Professor of Computer Science.4
  8. Co-authored "Locating and Editing Factual Associations in GPT" (ROME).3
  9. Became Principal Investigator and Director of the National Deep Inference Fabric.41
  10. Presented "NNsight and NDIF" at ICLR 2025.42

Other documented roles

Roles the sources document that are not confirmed as current.

  • Led the team that created Google Talk; contributed to Google Image Search, Google1
  • Member of the Trident team for Internet Explorer 4, Microsoft1
  • Postdoctoral Fellow in Martin Wattenberg's lab (2021–2022), Harvard University4

Honors and recognition

  • Sloan Research Fellowship · Alfred P. Sloan Foundation, 202541
  • MIT EECS Great Educators Fellowship · MIT Department of Electrical Engineering and Computer Science, 20154
  • NSF Graduate Research Fellowship · National Science Foundation, 19924

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

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

  1. David Bau: Interpretation of Deep Networks (Bau Lab, Northeastern University)David Bau · Official page · accessed Oct 21, 2026
  2. ndif-team/nnsight README and citation (MIT license)NDIF team (GitHub repository) · Repository · accessed Oct 21, 2026
  3. Locating and Editing Factual Associations in GPT (arXiv 2202.05262)arXiv · Paper · published Feb 10, 2022 · accessed Oct 21, 2026
  4. CV for David BauBau Lab, Northeastern University · Official page · accessed Oct 21, 2026
  5. National Deep Inference Fabric (NDIF) — homepageNDIF (Northeastern University) · Official page · accessed Oct 21, 2026