243rd RQC Seminar

• 講演者

  Dr. Linus Ekstrom
  ( Honda Research Institute Europe )

• 日程

  2025年12月9日(火), 16:00 - 17:00 (4:00p.m.-5:00p.m)

• 開催場所

  ハイブリッド（Zoom,
  Seminar Room 345-347, 3F, Main Research Building, Wako Campus / 研究本館3階 セミナー室 (345-347) (C01))

• 講演タイトル

  Improving and Scaling Quantum Multi Objective Optimization

• お問合せ

  norilab_rqc_assist[at]ml.riken.jp

講演概要
This talk explores how variational quantum circuits can address conflicting objectives using Quantum Multi-Objective Optimization (QMOO). I show how a persistent Pareto archive together with dominated-solution substitution makes hypervolume progress steady and reliable, with no additional quantum overhead, by using samples more effectively and stabilizing the classical optimization loop. I provide a simple mapping from classical RMNK landscapes to quantum cost Hamiltonians and use tensor-network simulators to examine larger instances than are practical with brute-force statevectors. I use this testbed to study further enhancements of QMOO by structured parameter sweeps and ablation studies of mixer design, angle parameterizations, and light-weight strategies that preserve search breadth such as archive-aware mixing and warm starts across circuit depth. These additions are intentionally low cost and produce modest but consistent gains in terms of the maximally achieved hypervolume.  On small benchmarks, QMOO attains hypervolume competitive with NSGA-II and NSGA-III while exploring the trade-off surface in a different way. What matters most is the interplay of shots, circuit depth, and problem structure. To end the talk I will sketch out some next possible steps for future work in the same framework.