Focus Period lund 2026
Postdoctoral Researcher
Uppsala University (Sweden)
Hoomaan Maskan Hezaveh completed his PhD in Mathematical Statistics at Umeå University, where he worked under the supervision of Alp Yurtsever, funded by the Wallenberg AI, Autonomous Systems and Software Program (WASP). His research centered on acceleration mechanisms in first-order methods and on difference-of-convex programming for structured nonconvex optimization. During his doctoral studies, Hoomaan Maskan Hezaveh had the opportunity to visit Konstantinos Zygalakis at the University of Edinburgh and Suvrit Sra at the Technical University of Munich. Before his PhD, he earned an MSc in Telecommunication Engineering and Signal Processing under the supervision of Mohammad Hossein Kahaie. In Spring 2026, Hoomaan Maskan Hezaveh will begin a new position at Uppsala University, working at the intersection of optimization and machine learning with Prashant Singh.
Presenting: The Multi-Block DC Function Class: Theory, Algorithms, and Applications
In this talk, we present the Multi-Block DC (BDC) class, a broad class of structured nonconvex functions that admit a DC (“difference-of-convex”) decomposition across parameter blocks. Thisblock structure not only subsumes the usual DC programming, it turns out to be probably more powerful. Specifically, we demonstrate how standard models (e.g., polynomials and tensorfactorization) must have DC decompositions of exponential size, while their BDC formulation is polynomial. This separation in complexity also underscores another key aspect: unlike DC formulations, obtaining BDC formulations for problems is vastly easier and constructive. We illustrate this aspect by presenting explicit BDC formulations for modern tasks such as deep ReLUnetworks, a result with no known equivalent in the DC class. Moreover, we complement the theory by developing algorithms with non-asymptotic convergence theory, including both batch and stochastic settings, and demonstrate the broad applicability of our method through several applications.