Printable PDF
Department of Mathematics,
Department of Mathematics,
University of California San Diego
****************************
Math 278B: Mathematics of Information, Data, and Signals
Henry Pritchard
UC San Diego
Architectural Guarantees for Convergent Plug-and-Play Proximal Gradient Descent
Abstract:
In this talk, I will present convergence guarantees for plug-and-play proximal gradient descent (PnP-PGD) that hold by architectural design. Unlike prior work requiring explicit Lipschitz constraints during training, which are difficult to enforce in practice, this approach leverages Learned Proximal Networks (arXiv:2310.14344). LPNs are constructed as gradients of Input Convex Neural Networks (ICNNs), and this architecture guarantees convergence of the resulting PnP-PGD scheme, independent of the training procedure. I will derive sublinear convergence under mild assumptions on the forward model.
May 29, 2026
11:00 AM
APM 2402
Research Areas
Mathematics of Information, Data, and Signals****************************
