🔹 Tah Poe seminar series LIX(59) Talk no.3

Speaker: Laphas Premcharoen

Title:  GEO: Enhancing Combinatorial Optimization with Classical and Quantum Generative Models

Abstract: Combinatorial optimization is a leading candidate for practical quantum advantage, yet current quantum optimization strategies frequently require translating real-world problems into Polynomial Unconstrained Binary Optimization (PUBO) expressions, which introduces significant variable overhead. This seminar introduces the Generator-Enhanced Optimization (GEO) framework, a novel “black-box” solver that bypasses this limitation by leveraging the generalization capabilities of machine learning generative models. We will specifically examine TN-GEO, a quantum-inspired instantiation using Matrix Product States. The session will cover TN-GEO’s application to an NP-hard cardinality-constrained portfolio optimization problem, demonstrating how it outperforms traditional classical solvers and competes with state-of-the-art metaheuristics by generating highly valuable, previously “unseen” solution candidates.