Uso de inteligencia artificial para domesticar sistemas cuánticos

El aprendizaje automático impulsa el autodescubrimiento de pulsos que estabilizan los sistemas cuánticos frente al ruido ambiental.

Controlar la trayectoria de una pelota de baloncesto es relativamente simple ya que solo requiere la aplicación de fuerza mecánica y habilidad humana. Sin embargo, controlar el movimiento de sistemas cuánticos como átomos y electrones plantea un desafío mucho mayor. Estas diminutas partículas son propensas a sufrir perturbaciones que pueden hacer que se desvíen de su ruta prevista de formas inesperadas. Además, el movimiento dentro del sistema se degrada, lo que se conoce como amortiguamiento, y el ruido de factores ambientales como la temperatura deteriora aún más su trayectoria.

Para contrarrestar los efectos de la amortiguación y el ruido, los investigadores del Instituto de Ciencia y Tecnología de Okinawa (OIST) en Japón han encontrado una manera de utilizar la inteligencia artificial para descubrir y aplicar pulsos estabilizadores de luz o voltaje con intensidad fluctuante a los sistemas cuánticos. Este método pudo enfriar con éxito un objeto micromecánico a su estado cuántico y controlar de manera óptima su movimiento. La investigación fue publicada recientemente en la revista Búsqueda de revisión física.

La idea básica es lograr el control cuántico mediante la aplicación del agente de IA (izquierda). Por ejemplo, para enfriar la bola cuántica (roja) hasta el fondo en presencia de ruidos ambientales, el controlador de IA, basado en el aprendizaje por refuerzo, identificaría pulsos de control inteligentes (gráfico polar promedio). Crédito: OIST

Los objetos micromecánicos, que son grandes en comparación con un[{» attribute=»»>atom or electron, behave classically when kept at a high temperature, or even at room temperature. However, if such mechanical modes can be cooled down to their lowest energy state, which physicists call the ground state, quantum behavior could be realized in such systems. These kinds of mechanical modes then can be used as ultra-sensitive sensors for force, displacement, gravitational acceleration, etc. as well as for quantum information processing and computing.

“Technologies built from quantum systems offer immense possibilities,” said Dr. Bijita Sarma, the article’s lead author and a Postdoctoral Scholar at OIST Quantum Machines Unit in the lab of Professor Jason Twamley. “But to benefit from their promise for ultraprecise sensor design, high-speed quantum information processing, and quantum computing, we must learn to design ways to achieve fast cooling and control of these systems.”

The machine learning-based method that she and her colleagues designed demonstrates how artificial controllers can be used to discover non-intuitive, intelligent pulse sequences that can cool a mechanical object from high to ultracold temperatures faster than other standard methods. These control pulses are self-discovered by the machine learning agent. The work showcases the utility of artificial machine intelligence in the development of quantum technologies.

Quantum computing has the potential to revolutionize the world by enabling high computing speeds and reformatting cryptographic techniques. That is why many research institutes and big-tech companies such as Google and IBM are investing a lot of resources in developing such technologies. But to enable this, researchers must achieve complete control over the operation of such quantum systems at very high speed, so that the effects of noise and damping can be eliminated.

“In order to stabilize a quantum system, control pulses must be fast – and our artificial intelligence controllers have shown the promise to achieve such a feat,” Dr. Sarma said. “Thus, our proposed method of quantum control using an AI controller could provide a breakthrough in the field of high-speed quantum computing, and it might be a first step to achieving quantum machines that are self-driving, similar to self-driving cars. We are hopeful that such methods will attract many quantum researchers for future technological developments.”

Reference: “Accelerated motional cooling with deep reinforcement learning” by Bijita Sarma, Sangkha Borah, A Kani and Jason Twamley, 29 November 2022, Physical Review Research.
DOI: 10.1103/PhysRevResearch.4.L042038