Quantum machine learning beyond parametrized quantum circuits: Quantum Gaussian Processes

Wednesday, 10 Sep 2025
11:40 am
Aula 1.A1
Facultad de Ciencia y Tecnología
UPV/EHU, Leioa

Abstract: 

In this talk we will review the trajectory of quantum machine learning (QML)—from early proposals in the 1990s, through the dequantization wave around 2018, to recent critiques of variational QML approaches, including scaling barriers such as barren plateaus, proliferation of local minima, and classical simulability. Such analysis will reveal a fundamental truth behind QML: Thus far, we have tried building quantum learning models by slapping together the better and most advanced elements of quantum computing and classical machine learning. But this approach has failed us time and time again. As such, we propose a new path for QML, where we advocate going back to the basics of learning theory and instead using one of the simplest, most interpretable model: Gaussian Process (GPs). We will first show how certain quantum stochastic processes form genuine GPs, and we will then use the power of Bayesian statistics to efficiently solve learning and optimization tasks.

Reference: García-Martín, Diego, Martín Larocca, and M. Cerezo. "Quantum neural networks form Gaussian processes." Nature Physics (2025): 1-7. https://www.nature.com/articles/s41567-025-02883-z