Mach-Zehnder and Interference Get Enhanced by Reinforcement Learning

Photonic integrated technologies provide an outstanding platform for several areas of research, from fundamental tests of quantum mechanics to quantum simulation and quantum communication.

Recently, mid-scale circuits have already been applied to various tasks, most notably to realize quantum walks or Boson Sampling experiments.

The potential of these technologies to reach large-scale implementations is rooted in the unique features of single photons, such as mobility, high bandwidth and ease of manipulation.

In this direction, major obstacles are represented by the availability of sources and detectors with limited efficiency, as well as by an imperfect control over their reconfigurable optical evolutions.

However, while practical solutions can be engineered for the two former stages, the latter opens up a challenge when compact and dense integration is desired for high-precision tasks.

The research project MAZINGER will take up this challenge by bringing together analytical and numerical tools, in order to enhance single-photon evolutions in state-of-the-art photonic applications.

To this end, MAZINGER will employ well-established tools from machine learning, such as reinforcement learning algorithms and saliency maps, to cope with changing environments and non-ideal reconfigurable components, respectively.

To strengthen our research, the project involves a collaboration with a leading group in experimental photonics, with the goal of testing out and applying our findings on a high-precision test of quantum mechanics.

In particular, the employed numerical techniques will be solidly based on the general framework of multi-photon interference, which has been investigated, both theoretically and experimentally, by the key players of this project.

Eventually, MAZINGER will pave the way for self-optimized applications of single- and multi-photon quantum interference in integrated photonic circuits.