April 25, 2026
Why and how variational quantum computing is limited.
During my 3-year undergraduate research group tenure in quantum computing, I was enamored with quantum machine learning (QML). I had been listening to companies like Google and Microsoft push quantum AI, claiming massive speedups. When presenting at SC23 and SC24, with the former focused on convolutional neural networks (CNNs), my professor had reminded me each time of the limitations of QML as it stood.
Variational quantum computing (VQC) is a paradigm that converts a problem into an optimization task, and then uses a classical device to train a parameterized quantum circuit to solve the problem. Sounds familiar. The base idea is related to gradient descent in typical neural networks and how they slowly adjust their weights to minimizing a loss function.
To begin optimizing, VQCs must select a set of hyperparameters: