Projects

Baiying Xin, Qi Deng

2026 · arXiv:2610.21595 · Preprint

For a class of Hölder-smooth problems arising from regularized $\ell_p$ regression with $1<p<2$, we develop a smoothing-based first-order method with an $\mathcal{O}(\varepsilon^{-1/p})$ matrix–vector complexity and prove a matching $\Omega(\varepsilon^{-1/p})$ lower bound for deterministic algorithms.