- This event has passed.
Fall 2025 GRASP SFI: Deepti Ghadiyaram, Boston University, “Interpreting and Leveraging Generative Representations”
September 24 @ 3:00 pm - 4:00 pm
This was a hybrid event with in-person attendance in Levine 307 and virtual attendance…
ABSTRACT
Generating high-quality photo-realistic visual content is a thriving area of research. In this talk, I will focus not on the generation process, but on understanding how rich visual semantic information is represented within generative models. Specifically, I will present our work that uses mechanistic interpretability tools like k-sparse autoencoders to probe various layers and denoising timesteps of different diffusion architectures. Next, I will present how to leverage interpretable concepts to control the generation process towards more safer and creative outcomes.
Presenter
Deepti Ghadiyaram
Deepti is an Assistant Professor at Boston University in the Department of Computer Science and also a Member of Technical Staff at Runway. Her research interests are on topics pertaining to safe and interpretable computer vision, improving the robustness of generative multimodal models. Prior to joining Boston University, she was a Senior Research Scientist at Fundamental AI Research (FAIR) in Meta AI. She has served as a program chair for NeurIPS 2022 Dataset and Benchmarks track, hosted several tutorials and organized workshops and an area chair for several years at CVPR, ICCV, ECCV, ACCV, AAAI, and NeurIPS.