Artificial intelligence designs better quantum experiments

Posted Mar 7, 2016 by Michael Thomas
Quantum science is so difficult to grasp — even for scientists — that a doctoral student designed an algorithm to better design experiments.
Optical components configured to produce photons in a high-dimensional GHZ entangled state  in an ex...
Optical components configured to produce photons in a high-dimensional GHZ entangled state, in an experiment designed by the computer algorithm called Melvin.
M. Malik/Univ. of Vienna
It all started with Mario Krenn, a doctoral student at the University of Vienna, CNET reports. He was trying to induce a specific quantum state using mirrors and lasers, but he eventually realized at some point that he was just guessing at how to properly design the experiment. That's where Melvin was born.
Melvin is the algorithm Anton Zeilinger and his team came up with. Krenn found that when the algorithm ran overnight, he found a .txt file with the solution in the morning. Details of the algorithm are published in the journal Physical Review Letters.
"That was quite an exciting day," Krenn said in a statement.
Melvin uses several points of information to construct experiments. In Krenn's experiment, for example, it used the mirrors and lasers and desired quantum outcome, then runs through setups at random. Its artificial intelligence allows it to learn from its solutions and mistakes so it can build on experience.
The team says Melvin has already come up with several experiments that a human team likely wouldn't have dreamed up. In fact, many of these experiments — much like quantum mechanics — are difficult for the scientists to fully grasp.
"I still find it quite difficult to understand intuitively what exactly is going on," Krenn said.
The team gave Melvin a big test with Greenberger-Horne-Zeilinger states (GHZ), a quantum state in which three or more photons become entangled. It ended up creating 51 experiments that led to quantum entanglement, with one leading to the desired GHZ state.
Though this AI is impressive, there's still one problem — the human mind needs to be capable of understanding the experiments and results. Still, Melvin could be a useful way of developing crucial experiments to further develop quantum computing and cryptography.