Stanford physicists build an atom-and-photon network that boosts AI memory

[GhostRider90]

A Stanford-led team created a new type of spin glass made from ultracold atoms and photons and used it to increase the memory capacity of a simple artificial intelligence system beyond what a conventional Hopfield network can achieve. The system, called a quantum-optical spin glass, works as an associative memory, the kind of AI that recalls a complete memory from partial information, similar to recognizing a person's face from a blurred photograph, and the team published the results in Science

Using laser tweezers, researchers created an array of ultracold atomic gases, called Bose-Einstein condensates, trapped inside an optical cavity formed by curved mirrors, with each cluster of roughly 10,000 or more atoms behaving as a single super atom. Using this small network of up to 20 spins as an associative memory, the quantum-optical spin glass showed up to seven times greater capacity to hold and recall memories than a traditional Hopfield network with the same number of spins, and also displayed short-term plasticity, a phenomenon resembling how synaptic connections between neurons change during learning

Senior author Benjamin Lev of Stanford said the team can now make neural networks at the atomic level that adjust themselves in a way that resembles how researchers believe biological brains actually learn. The system currently requires extremely cold temperatures and a vacuum chamber, meaning more research is needed before it could scale toward practical applications. Curious what people think this kind of physical hardware advance means for AI memory specifically, could quantum-optical systems like this eventually offer real advantages over purely digital approaches to associative memory, or does the extreme cooling requirement make practical deployment a distant prospect

Quantum Optical Glass Aiphys.org