Quantum Computation for Optical Sensor Design
We aim to improve optical instruments in aerospace by globally optimising diffractive structures under manufacturing constraints and transforming thematic information extraction AI-supported to quantum machine learning algorithms.
In this project, we investigate new approaches for the improvement of optical instruments for aerospace applications. The calibration of high-resolution hyperspectral camera systems is carried out with components that are designed to generate diffraction patterns according to strict targets. Their efficient design under manufacturing conditions leads to hard optimisation problems and we are investigating whether these can be solved with hybrid quantum computing. Also, data aggregation and downstream algorithms for content reconstruction have so far only been performed with sub-optimal system parameters (bandwidth, resolution, accuracy and so on). We are therefore investigating methods for semi-automated and AI-supported code transformation and optimisation for quantum machine learning algorithms.
We have been successfully developing camera systems and spectrometers for aerospace, security and transport for many years. These highly complex instruments contain a large number of optical, mechanical, electronic and software components that have to be harmonised in detail and controlled during operation in order to guarantee high data quality, availability and reliability despite limited resources. In addition to high image quality, crucial information, for example in hazardous situations, must also be reconstructed and communicated faster than before. These strict requirements against the background of ever-increasing dimensionality in the sensor data require new technologies and calculation methods in information acquisition and instrument design in the long term.
We are reaching the limits of available approaches to instrument design, as each instrument must be optimised in terms of data quality and boundary conditions with regard to a large number of environment-dependent parameters. The decisive factor is currently provided by engineers’ and scientists’ experience. A prototypical issue here is the design of diffractive structures. Forward calculations of light propagation are possible with today’s computers, but an exact inverse optimisation of diffraction designs is no longer feasible in a globally optimal way, as a simultaneous consideration of many local optima would be necessary. A similar problem arises in the field of data processing where AI-based methods already have enormous application potential.
However, these data-driven techniques also require a complex and often unstable learning process frequently involving considerable amounts of training data. Theoretical guarantees and robustness are still difficult and generalisation properties are based on empirical results. The search for quasi-optimal AI models is becoming increasingly complex and in some cases can only be carried out with considerable effort. For many use cases, this will no longer be sustainable in the foreseeable future and disruptive technologies such as quantum machine learning will become increasingly relevant for this purpose. Techniques relating to semi-automated and AI-supported code transformation and hybridisation allow us to keep pace with the rapidly increasing dimensionality of the problems.