By Donna Lu
Machine learning is growing ever more sophisticated, thanks to algorithms which pit two artificial intelligences against each other. These algorithms, known as generative adversarial networks (GANs), have already been used to create art, crack encryption codes, and produce uncannily real pictures of faces and animals.
Researchers have now combined GANs with another hot technology: quantum computing. Luyan Sun at Tsinghua University in Beijing, China and his colleagues have created a GAN on a quantum circuit.
GANs are formed of two neural networks, the generator and discriminator. As you might expect, the generator generates data – a picture of a face, say. The discriminator is given both real data and the fake created by the generator, and must _gure out which is which. The two AIs go back and forth, and The researchers created a quantum GAN using both a quantum generator and discriminator. Unlike ordinary computers, which require data to be stored in binary digits (bits) of 0s and 1s, quantum computers use quantum bits, or qubits, which can store a mixture of 0 and 1 same time.
Their generator was trained to replicate quantum data produced by a microwave resonator. Eventually, the discriminator was unable to distinguish between true and generated data.
In theory, quantum computers have a speed advantage over their regular counterparts in solving certain problems, such as factoring large numbers, says team member Dong-Ling Deng. But as the technology currently stands, quantum computers aren’t yet able to achieve this edge, says Sun.
The researchers believe that GANs on quantum computers may also have such a speed advantage, but they still need to de_nitively demonstrate that this is the case. Such a discovery would “be a milestone for quantum machine learning,” says Sun.
Achieving this “quantum supremacy” is thought to require at least 50 qubits for the simplest kinds of problem, but the team’s study used a system with just a single qubit. “There is still a long way to go,” says Deng.
Journal reference: Science Advances , DOI: 10.1126/sciadv.aav2761