IonQ announced promising early results with its partner, GE Research, to explore the benefits of quantum computing for modeling multi-variable distributions in risk management.
Using a Quantum Circuit Born Machine-based framework on standardized, historical indexes, IonQ and GE Research, the central innovation hub for the General Electric Company, were able to effectively train quantum circuits to learn correlations among three and four indexes.
The prediction derived from the quantum framework outperformed those of classical modeling approaches in some cases, confirming that quantum copulas can potentially lead to smarter data-driven analysis and decision-making across commercial applications. A blog post further explaining the research methodology and results is available here.
Peter Chapman, CEO and President at IonQ, said in a statement: “While classical techniques face inefficiencies when multiple variables have to be modeled together with high precision, our joint effort has identified a new training strategy that may optimize quantum computing results even as systems scale. Tested on our industry-leading IonQ Aria system, we’re excited to apply these new methodologies when tackling real world scenarios that were once deemed too complex to solve.”
While classical techniques to form copulas using mathematical approximations are a great way to build multi-variate risk models, they face limitations when scaling. IonQ and GE Research successfully trained quantum copula models with up to four variables on IonQ’s trapped ion systems by using data from four representative stock indexes with easily accessible and variating market environments.
By studying the historical dependence structure among the returns of the four indexes during this timeframe, the research group trained its model to understand the underlying dynamics. Additionally, the newly presented methodology includes optimization techniques that potentially allow models to scale by mitigating local minima and vanishing gradient problems common in quantum machine learning practices. Such improvements demonstrate a promising way to perform multi-variable analysis faster and more accurately, which GE researchers hope lead to new and better ways to assess risk with major manufacturing processes such as product design, factory operations, and supply chain management.