Uploaded March 2025 | Updated November 2026, 3 hours ago
Computer Science Seminar Series
March 6, 2025
“Understanding and Enhancing Deep Neural Networks with Automated Interpretability”
Tamar Rott Shaham, Massachusetts Institute of Technology
Deep neural networks are becoming incredibly sophisticated; they can generate realistic images, engage in complex dialogues, analyze intricate data, and execute tasks that appear almost humanlike. But how do such models achieve these abilities? In this talk, Tamar Rott Shaham will present a line of work that aims to explain behaviors of deep neural networks. This includes a new approach for evaluating cross-domain knowledge encoded in generative models, tools for uncovering core mechanisms in large language models, and their behavior under fine-tuning. She will show how to automate and scale the scientific process of interpreting neural networks with the Automated Interpretability Agent, a system that autonomously designs experiments on models’ internal representations to explain their behaviors. Shaham will demonstrate how such understanding enables mitigating biases and enhancing models’ performance. The talk will conclude with a discussion of future directions, including developing universal interpretability tools and extending interpretability methods to automate scientific discovery.
Tamar Rott Shaham is a postdoctoral researcher at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory working with Antonio Torralba. Shaham earned her PhD from the Technion Faculty of Electrical and Computer Engineering, supervised by Tomer Michaeli. She has received several awards, including the 2019 International Conference on Computer Vision Marr Prize, a Google Women Techmakers Scholarship, an Adobe Research Fellowship, a Rothschild Fellowship, the VATAT Zuckerman Israeli Postdoctoral Scholarship, and the Schmidt Sciences Israeli Women's Postdoctoral Award.
Computer Science Seminar Series
March 6, 2025
“Understanding and Enhancing Deep Neural Networks with Automated Interpretability”
Tamar Rott Shaham, Massachusetts Institute of Technology
Deep neural networks are becoming incredibly sophisticated; they can generate realistic images, engage in complex dialogues, analyze intricate data, and execute tasks that appear almost humanlike. But how do such models achieve these abilities? In this talk, Tamar Rott Shaham will present a line of work that aims to explain behaviors of deep neural networks. This includes a new approach for evaluating cross-domain knowledge encoded in generative models, tools for uncovering core mechanisms in large language models, and their behavior under fine-tuning. She will show how to automate and scale the scientific process of interpreting neural networks with the Automated Interpretability Agent, a system that autonomously designs experiments on models’ internal representations to explain their behaviors. Shaham will demonstrate how such understanding enables mitigating biases and enhancing models’ performance. The talk will conclude with a discussion of future directions, including developing universal interpretability tools and extending interpretability methods to automate scientific discovery.
Tamar Rott Shaham is a postdoctoral researcher at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory working with Antonio Torralba. Shaham earned her PhD from the Technion Faculty of Electrical and Computer Engineering, supervised by Tomer Michaeli. She has received several awards, including the 2019 International Conference on Computer Vision Marr Prize, a Google Women Techmakers Scholarship, an Adobe Research Fellowship, a Rothschild Fellowship, the VATAT Zuckerman Israeli Postdoctoral Scholarship, and the Schmidt Sciences Israeli Women's Postdoctoral Award.










