David Bau

David Bau is an assistant professor in the Khoury College of Computer Sciences. His research aims to improve the interpretability and control of large-scale machine learning, with a focus on generative neural models in vision and natural language.

Bau’s research characterizes the internal mechanisms of large models with the goal of understanding their algorithmic structure and identifying ways to directly control and improve their behavior. His lab studies the processing and representation of factual and procedural knowledge within neural networks, investigating how editing such mechanisms can achieve goals that go beyond the traditional machine-learning objective of imitating an empirical training distribution. Bau’s research work has led to insights in how direct model editing can be used to erase undesirable behaviors, edit and update incorrect factual knowledge, and reduce societal biases that are often amplified by large-scale machine learning.

Bau also directs open-source the open-source NNsight project, that aims to provide software infrastructure to crack open the black box of AI by enabling the development of transparent and widely-accessible interpretable methods for applying large-scale machine learning. Additionally, he serves as director for Pencil Code, an educational community aimed at making high-quality computer science education accessible to all K-12 students.

Bau is author of Numerical Linear Algebra, a widely-used textbook on numerical methods and matrix algorithms. In addition, he has published over 30 patents on a wide variety of software innovations, and has published over 30 referred papers on interpretable machine learning in venues including PNAS, NeurIPS, ICLR, SIGGRAPH, CVPR, ICCV, ECCV, and CoNNL.

Bau’s research is supported by Open Philanthropy. Before joining Northeastern, Bau worked at Google and Microsoft and helped lead the development of several widely-used products including Google Image Search and Microsoft Internet Explorer. He received his doctoral degree from Massachusetts Institute of Technology, his masters degree from Cornell University, and his undergraduate degree from Harvard University.