Multidisciplinary Institute in Artificial intelligence

The joys of Generative Models as Data Stores

ABSTRACT

"Rather than thinking of generative AI as intelligent agents, I argue it's more natural to consider it a type of effective data storage, where the data and its computational controls (search, retrieval, interpolation, etc) are co-located and co-dependent. Just as the Internet revolution was driven by Data but enabled by effective search, the AI revolution is still driven by Data enabled by generative AI. In this talk I will present a few recent results, where we use generative AI (primarily diffusion models) to enable various visual tasks, including visual data mining, counterfactual scene understanding, online 3D modeling, and interpolation in model weight space." 

SPEAKER

Alexei Efros is a professor at the EECS Department at UC Berkeley, where he is part of the Berkeley Artificial Intelligence Research Lab (BAIR). Before that, he spent a decade on the faculty of the Robotics Institute at CMU. He is also still remembered in lovely Oxford, where he did a post-doc with Andrew Zisserman. During 2007-2015, he has also been closely collaborating with Team WILLOW at École Normale Supérieure / INRIA in beautiful Paris. 

RESEARCH 

The central goal of his research is to use vast amounts of unlabelled visual data to understand, model, and recreate the visual world around us. His research has been mainly in data-driven computer vision, as well as its projection onto computer graphics and computational photography. In the last five years, his lab has been at the forefront of reviving self-supervised learning. Other interests include human vision, visual data mining, robotics, and the applications of computer vision to the visual arts and the humanities. 

Link website