Data-driven Learning and Control - DDLC

Abstract: Can we build neural architectures that go beyond Transformers by leveraging principles from dynamical systems? In this talk, I will introduce a novel approach to sequence modeling that draws inspiration from the emerging paradigm of online control. The method is based on a new algorithm for learning in nonlinear dynamical systems, to achieve efficient long-range memory, fast inference, and provable robustness. I will present theoretical insights, empirical results. and recent advancements in fast sequence generation and provable length generalization.