Artificial Intelligence came up with new solutions to quantum experiments. (Image Credit: Pixabay)
Mario Krenn, a Quantum physicist, developed a machine-learning algorithm named MELVIN, purposed to discover new solutions to quantum experiments while mixing and matching its building blocks. Along the way, MELVIN discovered some intriguing solutions, but one of them didn't make sense. In this context, MELVIN solved the problem of building extremely complex entangled states involving multiple photons. What's strange about this is that Krenn, Anton Zeilinger of the University of Vienna, and their colleagues didn't implement any parameters for MELVIN to produce these complex states. He realized that the machine learning algorithm rediscovered an experimental arrangement that was created in the 1990s.
Since then, various teams have performed experiments discovered by MELVIN, enabling them to test quantum mechanics using new methods. Krenn and University of Toronto colleagues improved their machine learning algorithms. Their latest development, an AI called THESEUS, is faster than MELVIN by orders of magnitude. Although it takes days or weeks to understand what MELVIN produces, THESEUS can be understood right away.
Krenn's team wanted to achieve a 3D GHZ state that was a superposition of states 000, 111, and 222. These states were crucial toward secure quantum communications and faster quantum computing. In 2013, the team designed experiments and performed calculations to see if it could produce quantum states, but to no avail. During its initial development, MELVIN calculated an experimental setup's output. Afterward, Krenn improved the program so that it could add lasers, nonlinear crystals, beam splitters, phase shifters, holograms, etc. This provided MELVIN with the ability to perform calculations that experimentalists use on an optical bench. Then, the program mixed and matched the building blocks and performed the calculations, which produced the result.
Krenn ran an experimental toolbox containing two crystals capable of producing two photos entangled in three dimensions. He expected MELVIN to discover configurations that mixed these photons to produce entangled states of nine dimensions. Instead, it found a higher entanglement solution by using a technique developed in 1991. However, the solution involved using two crystals, which were placed inside an interferometer. Afterward, Krenn realized that MELVIN's setup involved more than two crystals that produced pairs of photons, which caused the detectors to overlap their paths.
In March, researchers at the University of Science and Technology in China collaborated with Krenn to fabricate the setup on a photonic chip and perform experiments. They collected over 16 hours' worth of data, which was only possible due to the chip's optical stability.
While attempting to figure out what MELVIN generated, the team discovered that the solution mirrored graphs. The quantum experiments showed that every path a vortex took was represented by a vertex. Meanwhile, a crystal was represented by an edge connecting two vertices. At first, MELVIN generated this graph and performed mathematical calculations on it. This involves producing a similar graph where each vertex connects to one edge, making it easier to calculate the final quantum state.
So, it made a test… but could AI take my tests for me?
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell
Top Comments