Seminar: Building Quantum Engines for Large-Scale Decision Making
Jiaqi Leng
Postdoc Fellow
Simons Institute for the Theory of Computing
University of California, Berkeley
Friday, February 6, 2026
11:00a.m. - 12:15 p.m.
Academic Building One, Room 3740
Abstract
Decision making involves selecting high-quality actions from complex data and sits at the core of modern machine learning, Al, and robotics. In practice, these tasks often reduce to optimization or control problems in high-dimensional, noisy, and possibly non-convex landscapes, where classical methods can struggle with poor scaling and limited accuracy. In this talk, I present an end-to-end methodology for building quantum solutions that couples provable algorithmic speedups with hardware-efficient implementations. First, for optimization, I introduce Quantum Hamiltonian Descent (QHD), a quantum analogue of gradient descent that leverages quantum tunneling and achieves exponential speedups in structured non-convex regimes. I then describe QHDOPT, an open-source software that enables deployment of QHD on both gatebased and analog quantum backends. Second, for optimal control, I develop a differentiable hybrid quantum-classical framework that combines a quantum gradient estimator with a classical optimizer. Based on a novel quantum subroutine for solving Lyapunov equations, this framework yields a super-quadratic speedup for large-scale linear control problems.Together, these two case studies represent two complementary routes to provable and useful quantum advantage for decision making. No prior background in quantum mechanics is assumed.
Biography
Jiaqi Leng is a Postdoc Fellow at the Simons Institute for the Theory of Computing and the Math Department at UC Berkeley, mentored by Umesh Vazirani and Lin Lin. He received his Ph.D. from the University of Maryland in 2024. His research focuses on developing efficient quantum algorithms and supporting software for large-scaleoptimization, machine learning, and scientific computing. His work has been recognized by leading journals and conferences, including PNAS, QIP, NeurlPS, and ICML.