Explore the emerging paradigm of implicit models in deep learning through this seminar presented by Alicia, a Ph.D. candidate at UC Berkeley. Dive into the concept of "equilibrium" equations that define predictions, contrasting with traditional feedforward structures. Examine the potential of incorporating loops in neural networks to better mimic brain function and enable complex higher-level reasoning. Investigate the challenges of well-posedness in implicit models and learn about various training problems, including a convex optimization approach. Discover connections to architecture optimization, model compression, and robustness. Gain insights into experimental results showcasing the potential of implicit models in parameter reduction, feature elimination, and mathematical reasoning tasks.
Overview
Syllabus
[Seminar Series] Implicit Deep Learning
Taught by
VinAI