A0998
Title: A derivative-free approach for parameter inference in hidden quantum Markov models
Authors: Ning Ning - Texas A&M University (United States) [presenting]
Abstract: Hidden quantum Markov models (HQMMs) provide a quantum-inspired framework for modeling complex sequential data, offering greater expressive power than classical Hidden Markov models (HMMs). Existing learning algorithms for HQMMs typically rely on gradient-based optimization of the log-likelihood function, which can be computationally intensive and sensitive to local minima. The aim is to propose a new and general method for inferring HQMM parameters that avoids the computation of derivatives of the log-likelihood. The approach is broadly applicable to HQMMs with arbitrary Kraus operator structures and enables learning in settings where differentiability is difficult to guarantee or gradient computation is costly. The proposed algorithm is validated on synthetic datasets, demonstrating that it can successfully recover HQMM parameters, and baseline methods are outperformed in both accuracy and computational efficiency.
