EUROPT 2025
Abstract Submission

497. Blockwise DC Programming

Invited abstract in session TC-6: Structured nonsmooth optimization -- Part I, stream Nonsmooth and nonconvex optimization.

Tuesday, 14:00-16:00
Room: B100/7013

Authors (first author is the speaker)

1. Hoomaan Maskan
Umeå University
2. Pouria Fatemi
Technical University of Munich
3. Alp Yurtsever
Umeå University
4. Suvrit Sra
Massachusetts Institute of Technology

Abstract

Difference of convex (DC) programs are a class of structured non-convex problems. Due to their general form, many applications find their abstractions through the lens of DC programming. In this work, we target the class of blockwise DC (BDC) functions. Careful formulation reveals the broad applicability of these programs in, for example, training deep neural networks. We propose blockwise DC algorithms (BDCA) to treat these problems under various conditions, such as generalized smoothness and stochastic gradients.

Keywords

Status: accepted


Back to the list of papers