Current Projects
I am currently working on researching universally optimal methods for composite minimization. Specifically, we want to allow for heterogeneous components that vary in their level of smoothness (from standard $L$-smooth functions with Lipschitz gradients to nonsmooth Lipschitz functions themselves). We further look for where we can benefit from heterogeneous levels of uniform convexity in the components. Future work entails characterizing these dual notions and applying them to novel research on interpolation theorems and performance estimation.
Papers (preprints)
- A Universally Optimal Primal-Dual Method for Minimizing Heterogeneous Compositions arXiv