About this event
Hürtig Hall, 130
free event
free event
How can machines learn as efficiently as humans and animals?
How can machines learn to reason and plan?
How will machines learn representations of perception and action plans at multiple levels of abstraction, enabling them to reason, predict, and plan on multiple time horizons?
In this seminar, Yann LeCun proposes a path to autonomous intelligent agents based on a novel modular cognitive architecture and a somewhat novel self-supervising training paradigm. Central to the proposed architecture is a configurable predictive world model that agents can plan. Behavior and learning are driven by a unique set of differentiable cost functions. The world model uses a new type of energy-based model architecture called H-JEPA (Hierarchical Joint Embedding Predictive Architecture). H-JEPA learns a hierarchical abstract representation of the world that is both maximally informative and maximally predictable. The corresponding working paper is available here.
The lecture will be followed by a chat with Yann LeCun and Usama Fayyad, Executive Director of the Institute for Experiential AI.
