The exponential increase in computing power that is taking shape opens up new perspectives in terms of machine learning. But without revolutionizing the concept.

With the advent of the quantum computer, computing will enter a new era. This future generation of supercomputer will double the computing power with each new qubit (or quantum bit). The promise ? To be able to process in a few seconds processing that takes thousands of years for traditional supercomputers. But to achieve this result, it will first be necessary to create sufficiently stable qubits. What is not yet won. Amazon, Google, IBM, Microsoft, as well as Atos in France are in the race. At the same time, everyone is already anticipating the next step: the development ofalgorithms optimized for the new architecture. And among these applications, artificial intelligence figures prominently.

“Thanks to the power provided by quantum computing, it will become possible to train machine learning models on gigantic learning bases”, underlines Florian Carrière, senior manager in charge of emerging technologies at Wavestone. From there, a quantum algorithm based on a probabilistic method could identify connections that were once impossible to discern. “An automatic language processing model like GPT 3 already has 175 billion learning parameters. We will be able to go much further. This will make it possible to overcome a new gap in terms of complex translation or sentiment analysis. “says Cyril Allouche, in charge of the quantum R&D program at Atos.

## Billions of billions of parameters

Same for computer vision. In this field, convolutional neural networks will be able to process billions of billions of parameters, benefiting in particular from reentrant networks. Recognition of shapes and scenes, in the autonomous vehicle for example, will reach an unprecedented degree of finesse and precision. Alongside the size of quantum memory, the benefit will, of course, be in accelerating the speed of learning.

“Using quantum entanglement to create a new type of neural network has no advantage”

Beyond the use of existing artificial intelligence models, could quantum infrastructures give rise to new kinds of learning structures? “Using quantum entanglement to create a new type of neural network, in which the weights would not be real values but superimposed values, does not present any advantage. Both researchers and industrialists have reached a consensus on this point”, replies Cyril Allouche.

As for the models of machine learning existing ones, they will be easily applicable to a stable quantum environment with a minimum number of qubits. “Most of the research published on the subject aims to convert current learning algorithms in order to run them on quantum environments”, insists Xavier Vasques, CTO and distinguished data scientist at IBM France. “Generative Adversarial Networks or GANs or Support Vector Machines (SVMs) are good examples.”

In the case of SVMs, kernel functions (exponential, Gaussian, hyperbolic, angular, linear) solve classification, regression or anomaly detection problems within 2D or 3D space. “The greater the volume of characteristics, the more expensive the calculation will be in terms of machine power. We therefore gain by transforming these functions into quantum algorithms to speed up the calculation”, explains Xavier Vasques. Another advantage of quantum computing: it makes it possible to more quickly define the hyperparameters of a model in a three-dimensional space by parallelizing the calculations in 3D.

“Quantum algorithms manage to detect patterns in noisy data that classical algorithms do not identify”

“The first results that we have obtained on quantum algorithms show that they manage to detect patterns in noisy data where classical algorithms do not identify any”, observes Xavier Vasques. “To observe the production of the Higgs boson which is extremely faint, CERN uses quantum SVMs to detect telltale micro-events in the big data produced by its particle accelerator. These support vector machines generate a classification of the signal, the background noise…” Quantum SVMs that achieve the same result as the classifiers developed by CERN openlab based on classical methods. “This suggests great progress as research in quantum hardware progresses”, concludes Xavier Vasques.

## NISQ AI

From SVM to convolutional neural networks, quantum computing promises to accelerate research in multiple fields: genetic analysis, protein structure prediction and discovery of new treatments, logistics optimization, product recommendation engine, detection of fraud, chemistry, development of new materials… In industry, Airbus and even EDF are involved in research into quantum computing. Another highly invested sector: finance. In this area, Barclays, Goldman Sachs and JPMorgan are in the running. Their objective is in particular to take advantage of statistical quantum machine learning to arrive at more robust prediction models or to refine the scoring of assets or credits.

While waiting to benefit from stable quantum machines, the first NISQ systems (for Noisy Intermediate-Scale Quantum) should see the light of day by 2023. These are quantum machines with between 50 and a few hundred qubits, but with stability too short to perform certain operations. “SVMs, for example, will not work in NISQ mode. Reinforcement learning is more promising. As for neural networks, they will require hundreds of thousands or even millions of qubits to work in this mode, the encoding cost being quadratic per compared to the number of parameters to be optimized”, explains Cyril Allouche at Atos. Quantum deep learning could nevertheless prove more robust to the noise of NISQ systems in image recognition. As often in computer science, the first quantum AIs should not be without constraints.

We wish to give thanks to the author of this write-up for this outstanding web content

What quantum computing will change in AI

Explore our social media accounts and other pages related to themhttps://www.ai-magazine.com/related-pages/