Adaptive first-order methods
I study how algorithms can exploit unknown smoothness and growth conditions, with a focus on structured nonsmooth and function-constrained optimization.
Complexity guarantees · Bundle & level methods
Optimization · Theory & Computation
Zhenwei Lin.
Postdoctoral Researcher Purdue University
I develop first-order methods that adapt to unknown problem structure, and build scalable solvers for large-scale optimization.
At Purdue, I work with Zhe (Jimmy) Zhang. Previously, I received my Ph.D. from Shanghai University of Finance and Economics, advised by Qi Deng.
Building better algorithms.
01 / Research
From structure to algorithms.
Optimization at scale
I develop practical solvers for large-scale conic and semidefinite programs, connecting first-order algorithms with low-rank structure and GPU computation.
Conic programming · Low-rank methods · GPUs
02 / Selected work
Recent ideas & results.
2026
Preprint
Accelerated Prox-Level Methods for Unknown Piecewise-Smooth Optimization II: Function-constrained Optimization
Restarted Penalty APEX handles unknown piecewise-smooth objectives and constraints, with certificates for both optimality and feasibility under quadratic growth.
2026
Preprint
Accelerated Prox-Level Methods for Unknown Piecewise-Smooth Optimization I: Convex Optimization
APEX adapts to unknown piecewise smoothness under quadratic growth, with improved oracle complexity and a verifiable termination certificate.
2026
Preprint
Global Linear Convergence of the Proximal Bundle Method under Unknown Piecewise Smoothness and Quadratic Growth
Explaining how retaining sufficiently many cutting planes enables linear convergence for convex piecewise-smooth objectives under quadratic growth.
2026
INFORMS Journal on Computing / Accepted
Uniformly Optimal and Parameter-free First-order Methods for Convex and Function-constrained Optimization
Optimal oracle complexity for convex function-constrained optimization, without requiring prior knowledge of smoothness parameters.
2025
Preprint
A Practical GPU-Enhanced Matrix-Free Primal-Dual Method for Large-Scale Conic Programs
PDCS and cuPDCS combine matrix-free primal-dual methods with GPU computation for large-scale conic programs.
Get in touch
Let’s talk optimization.
For research discussions and potential collaborations.