Quantifying the benefits of quantum AI systems
We are making the complete end-to-end pipeline of quantum artificial intelligence (AI) systems comparable, quantifying any quantum advantages over classical methods, developing a demonstrator, and thus initiating the process of standardisation in the field of quantum AI systems.
Although interest in AI methods on quantum hardware for practical applications is steadily increasing, the appropriate algorithms are still in their infancy. Feasibility studies have so far not been able to demonstrate significant advantages offered by quantum systems beyond specialised constructed problems. With Quant²AI, we want to evaluate the possible advantages of quantum AI systems heuristically using a reproducible method of comparison. This benchmark would enable developers to test their methods in a standardised way and helps users to identify suitable algorithms for their field of application.
The performance of quantum computers has increased dramatically in the last three years. US technology company Google has even announced that they have achieved quantum superiority or quantum advantage. This performance increase raises the expectation for the implementation of complex AI algorithms on quantum hardware and the realisation of advantages for practical applications over classical hardware in the medium term. Operating a quantum computer involves considerable effort, and this is unlikely to change in the foreseeable future. This raises the question of whether non-classical hardware can ever offer more efficient solutions. Even if the implementation of an AI method includes sub-algorithms whose quantum advantage has been demonstrated, this advantage may be lost again when considering the overall system. This could be due to, for example, the need to encode classical data into quantum states or to perform a measurement in order to read out the results. A realistic assessment of quantum advantages must therefore take the complete process chain into account in order to counteract any exaggerated benefits of sub-processes.
Our benchmark aims to compare algorithms in an unbiased and reproducible way by using statistical methods and hyperparameter searches. In each case, we compare the optimal configuration of a specific AI process chain. A modular structure allows us to identify the origin of an advantage and to find a starting point for further investigation.
We see applications for our benchmark beyond research, primarily in an industrial context. Here, there is a need to evaluate the use of quantum AI systems for in-house applications and to be able to estimate the time frame in which suitable quantum hardware for performing these algorithms will be available. To do this, we are designing the benchmark to be accessible not only to a company’s developers but also to its decision-makers. Technically savvy decision-makers can thus make informed decisions about the use of quantum AI systems. In this way, we are helping quantum technologies move beyond the current ‘gold rush’ mentality and towards greater investment security.