We are building a software infrastructure for hybrid Quantum High-Performance Computing (Q-HPC) and developing, using and evaluating hybrid algorithms for anomaly detection in combination with causal inference.
Our topic is Q-HPC, which combines elements of classical HPC with elements of quantum computing. Such hybrid approaches are of great practical importance because quantum computing is not superior to classical computing for all tasks. In this project, we are applying hybrid Q-HPC in the field of anomaly detection and testing and analysing it using large quantities of telemetry data. In addition, we are integrating the methodology of causal inference, which deals with the data-driven analysis of cause-effect relationships.
The claimed superiority of quantum computing over classical computing can so far only be demonstrated in practice for selected problems. Moreover, the computing power of current quantum hardware is still limited due to low qubit numbers and high error rates. As computing hardware becomes increasingly diverse, it will likely become possible to integrate quantum computers into heterogeneous computing clusters in the future. The quantum computers of today’s noisy intermediate-scale quantum (NISQ) era, for example, can already be profitably deployed in this way and become more attractive to a broader user base in hybrid form.
In the project, we also use causal inference. This is a modern research field within statistics and machine learning and deals with the data-driven analysis of cause-effect relationships. It thus enables an understanding of complex processes that goes beyond mere statistical dependencies, even when controlled experiments are not possible.
The industrial use case for hybrid computing we are investigating is the automated detection of anomalies in telemetry data. Based on a classical computing infrastructure for data analysis, we are examining the benefits of hybrid computing in terms of computational speed and accuracy for this use case.
Causal inference methods are particularly useful for this use case because there are a large number of individual measured variables and the selection of a relevant subset for data analysis is typically based on predefined expert knowledge. One challenge will be to use causal inference to improve the selection process based on incomplete prior knowledge.