Trainability of a quantum-classical machine in the NISQ era

Advancements in classical computing have significantly enhanced machine learning applications, yet inherent limitations persist in terms of energy, resource and speed. Quantum machine learning algorithms offer a promising avenue to overcome these limitations but bring along their own challenges. This experimental study explores the limits of trainability of a real experimental quantum classical hybrid system implementing supervised training protocols, in an ion trap platform. Challenges associated with ion trap-coupled classical processor are addressed, highlighting the robustness of the genetic algorithm as a classical optimizer in navigating complex optimization landscape inherent in binary classification problems with many local minima. Experimental results, focused on a binary classification problem, reveal the superior efficiency and accuracy of the genetic algorithm compared to gradient-based optimizers. We intricately discuss why gradient-based optimizers may not be suitable in the NISQ era through thorough analysis. These findings contribute insights into the performance of quantum-classical hybrid systems, emphasizing the significance of efficient training strategies and hardware considerations for practical quantum machine learning applications. This work not only advances the understanding of hybrid quantum-classical systems but also underscores the potential impact on real-world challenges through the convergence of quantum and classical computing paradigms operating without the aid of classical simulators.