Conceptual understanding through efficient inverse-design of quantum optical experiments

The design of quantum experiments can be challenging for humans. This can be attributed at least in part to counterintuitive quantum phenomena such as superposition or entanglement. In experimental quantum optics, computational and artificial intelligence methods have therefore been introduced to solve the inverse-design problem, which aims to discover tailored quantum experiments with particular desired functionalities. While some computer-designed experiments have been successfully demonstrated in laboratories, these algorithms generally are slow, require a large amount of data or work for specific platforms that are difficult to generalize. Here we present Theseus, an efficient algorithm for the design of quantum experiments, which we use to solve several open questions in experimental quantum optics. The algorithm' core is a physics-inspired, graph-theoretical representation of quantum states, which makes it significantly faster than previous comparable approaches. The gain in speed allows for topological optimization, leading to a reduction of the experiment to its conceptual core. Human scientists can therefore interpret, understand and generalize the solutions without performing any further calculations. We demonstrate Theseus on the challenging tasks of generating and transforming high-dimensional, multi-photonic quantum states. The final solutions are within reach of modern experimental laboratories, promising direct advances for empirical studies of fundamental questions, as well as technical applications such as quantum communication and photonic quantum information processing. In each case, the computer-designed experiment can be interpreted and conceptually understood. We argue that therefore, our algorithm contributes directly to the central aims of science.