Developer suite for automated machine learning with QC
Quantum computing enables an acceleration of machine learning (ML) approaches and the development of new solution approaches. However, the implementation of such quantum algorithms is use case specific and therefore requires characteristic development by interdisciplinary quantum software engineers. The AutoQML project aims to simplify these processes. The goal is to extend approaches for the automation of machine learning by quantum computing, in order to be able to, among others, solve problems in the production and automotive area more easily and faster.
Challenge and innovation
A similar development can be expected in the field of quantum computing (QC): In general, and explicitly in quantum machine learning (QML), quantum computing promises to solve certain problems many orders of magnitude faster than is possible using conventional computing power. However, a key challenge remains the shortage of skilled workers. To enable a broad industrial application of QML, a high degree of user-friendliness and automation is indispensable.
The AutoQML project aims at both solving challenges of current AutoML approaches and adapting the successful approach of AutoML to simplify access to QML methods for future users.
Specifically, the project should:
1. support the automatic integration of quantum components into today's ML solutions in order to exploit the performance, runtime and complexity advantages of quantum algorithms in an industrial context,
2. provide conventional developers with easy access to both classical and quantum ML algorithms, including hyperparameter optimization,
3. provide (Auto)ML experts with quantum solutions in order to design hybrid overall solutions that make use of the best components from both worlds in an application-specific way,
4. use quantum solutions for the AutoML process to reduce the high hardware requirements of large HPC systems, thus making existing AutoML processes more economical and ecological.
In the AutoQML project, quantum solutions for optimal ML algorithm selection and hyperparameter optimization are developed, and state-of-the-art QML algorithms are integrated into the existing ML pool of the AutoML framework. The key approaches are:
1. Use-case specific QML algorithms are developed, simulated and demonstrated on real quantum backends under the given restrictions.
2. The developed QML algorithms will be automated and integrated into the existing AutoML framework.
3. The tools and methods developed in the project will be integrated into the PlanQK platform as an open-source solution and thus made available to (quantum) developers.
4. Quantum optimization algorithms for efficient and appropriate (Q)ML algorithm selection, incl. hyperparameter optimization supported by them, will be developed.
January 2022 – December 2024
Fraunhofer IAO (project lead), Fraunhofer IPA, GFT Integrated Systems, USU, HQS Quantum Simulations, IAV, KEB, Trumpf, Zeppelin
Total costs: € 7.0 million
Funding volume: € 4.5 million