WQCG: Expressivity vs. Stability in Quantum Model Development

Speaker: Jacob Cybulski, founder of Enquanted

Abstract

„Make your quantum model more expressive!” This has become a familiar refrain in response to anyone seeking advice in quantum machine learning communities. Yet the notion of expressivity is as fashionable as it is elusive, particularly in the quantum setting.

In this talk, we will explore what expressivity really means and how it manifests itself in quantum model development. Is expressivity primarily a consequence of increasing model complexity, or does it emerge from the finesse of the modelling approach?

Should we always pursue greater expressivity, and if not, where should we stop? Building on my recent investigations into quantum model expressivity and its relationship with trainability, I will discuss the delicate balance between these two often competing objectives. However, a third dimension has emerged from this analysis: model stability. Through theoretical considerations and empirical observations, I will explore how expressivity, trainability, and stability interact, and why this relationship may be crucial for developing reliable and effective quantum machine learning models.

The talk will provide both conceptual perspectives and practical insights, aiming to stimulate discussion among researchers at different stages of their engagement with quantum modelling.