We have been looking for support for our sensor project QCOptSens | CrossCompiler. HQS Quantum Simulations has now been awarded the contract. The start-up will develop a cross-compiler for the automated transformation of a machine learning problem for optical sensing into a QML pipeline. The subcontractors are the Fraunhofer Institutes for Industrial Engineering IAO and for Manufacturing Engineering and Automation IPA.
Optical sensors are important tools in aerial and satellite image recognition for hazard defence and disaster control. Increasingly, multi-spectral and hyperspectral data is being used to enable detailed analyses of land surfaces, the sea and the atmosphere. For example, materials that appear identical in the simple RGB colour space can be distinguished.
However, automatically analysing this data is very time-consuming due to the increased volume of data. Quantum machine learning (QML) creates new approaches, but requires new knowledge and skills in automated image recognition due to the variety of data and systems: the most suitable algorithms and approaches must be selected from the many possible and optimised for the respective problem. This is especially true if the methods are to be used on different, real quantum computers and the respective hardware conditions must be taken into account when designing the code.
The best method for the right machine
In classic machine learning, this challenge of selecting the right methods is solved by automating the creation of a machine learning pipeline: Automated Machine Learning, AutoML. In the field of quantum machine learning, the two Fraunhofer Institutes IPA and IAO have developed an AutoML framework for the end-to-end automation of QML algorithms as part of the BMWK-funded AutoQML project.
For QCOptSens, HQS Quantum Simulations – with Fraunhofer IAO and IPA as subcontractors – will develop a cross-compiler based on this framework that enables the automated transformation of a user-defined ML problem from the field of optical sensing into a QML pipeline. The AutoQML framework provides a good basis for this because it automatically selects the best method from an (expandable) pool of algorithms with regard to a given metric and budget. The necessary pre-processing steps and the search for suitable hyperparameters are also carried out and optimised.
Collaboration in the QCOptSens project
With the call QCOptSens | CrossCompiler (Automated Code Transformation for Quantum Machine Learning), we were looking for companies to investigate new techniques for semi-automated and AI-supported code transformation and hybridisation and to assess their potential. Research and development in this area is essential to keep pace with the rapidly increasing dimensionality and complexity of data and algorithms in the field of quantum computing-based calculations in the near and longer term. Methods will be proposed, implemented and evaluated using examples from the field of thematic content reconstruction with hazard context.
HQS Quantum Simulations
The start-up from Karlsruhe specialises in the development of modern software applications for the simulation and analysis of materials at quantum level. With its solutions, it aims to open up new possibilities for accurately and efficiently predicting and analysing materials.
