Quantum computers can augment AI

Quantum computers can augment AI

AI Business

September 17, 2019

3 Min Read

by Zapata Computing 17 September 2019

Generating data – images or videos – has been an important element of AIresearch. After all, teaching a machine to come up with new images not onlyendows the android with a sense of creativity – precious in the quest forArtificial General Intelligence – but also opens the door to many businessapplications. Generative models are however hard to train and to deploy, but theycould be improved through quantum computing.

Quantum computers (QC) are small today and are not likely to replacemost AI/ML algorithms in the near future because they simply cannot process themassive amounts of data that are required to train useful models. For instance,QCs will not likely have much of an impact on discriminative models, as thetraining data is typically a massive set of classically encoded data (though thereis much potential in areas like chemistry, where thedata is coming from a quantum system). 

We are, however, discovering ample opportunities for QCs to augmentexisting models.  Our insight at Zapata Computing is to rely onthe quantum computer only for the hardest computational subtask, and allow the classicalcomputer to do the rest.  In particular,one opportunity of a quantum computer comes from the fact that they are able togenerate probability distributions that classical computers cannot efficientlygenerate. In fact, within the next year or two, we will likely see anannouncement of so-called “quantum supremacy”, where a quantum machine is ableto quickly generate complex distributions that no classical computer anywherecould replicate.

One such hybrid system that could be acceleratedis a variant of a Generative Adversarial Network (GAN). A GAN is a neuralnetwork that works in a way analogous to counterfeiting; one part of thenetwork (the “generator”) acts like a money forger and the other part (the“discriminator”) tries to sort out the fake money from the real money. GANshave been particularly successful in image generation, with applications fromastrophysics to fashion modeling. One limitation of GANs is that discriminatorstend to improve much faster than generators, and this slows down the trainingprocess.

An associate adversarial network (AAN) is a variant of a GAN which tries to improve onthe model by introducing a kind of eavesdropper into the mix—more precisely, using a Restricted Boltzmann Machine (RBM) tofeed some information from the discriminator to the generator, speeding up itslearning. One opportunity, then, is to use a Quantum Boltzmann Machine toreplace this classical component.  Asthis part of the neural network has a relatively small number of neurons, it ispossible to leverage quantum computing to accelerate this subtask and speedupthe overall training time of the AAN.

As quantum computers grow in size, speed, andcapability, we envision more opportunities and faster improvements. We feelcertain there will come a time when AI is increasingly reliant on quantumcomputing, taking on a complementary role similar to the role of GPUs and FPGAstoday.

The eminent Carl Sagan once said, "We are a wayfor the Cosmos to know itself." Quantum computers will be key to takingthat knowledge to the next level.

Get the newsletter
From automation advancements to policy announcements, stay ahead of the curve with the bi-weekly AI Business newsletter.