IBM: the Advantages of the Quantum Computer Over Traditional Computers

Among the most promising applications of quantum computing, quantum machine learning is set to make waves. But how this might come to fruition remains a bit of a mystery.

IBM researchers now claim that they have mathematically proven that by using a quantum approach, some machine learning problems can be solved faster than with conventional computers.

The machine learning is an established branch of artificial intelligence , which is already used in many industries to solve various problems. This involves training an algorithm with large data sets, in order to allow the model to identify different patterns and ultimately calculate the best answer when presented with new information.

Quantum computing and machine learning

With larger data sets, a machine learning algorithm can be optimized to provide more accurate answers, but this comes at a computational cost that quickly reaches the limits of traditional devices. This is why the researchers hope that one day they will be able to take advantage of the enormous computing power of quantum technologies to take machine learning models to the next level.

One method in particular, called 'quantum nuclei', is the subject of numerous research articles. In this approach, the quantum computer intervenes only for a part of the global algorithm, by extending what is called the space of characteristics, that is to say the collection of characteristics used to characterize the data supplied to the model, such as 'sex' or 'age' if the system is trained to recognize patterns in people.

To put it simply, using the quantum nucleus approach, a quantum computer can distinguish between a greater number of features and, therefore, identify patterns even in a huge database, whereas a classical computer would not see. only random noise.

Classification problem

IBM researchers set out to use this approach to solve a specific type of machine learning problem called classification. As the IBM team explains, the most common example of a classification problem is a computer that receives photos of dogs and cats and needs to train with that dataset. The end goal is to allow him to automatically tag all future images he receives as either a dog or a cat, with the goal of generating accurate tags in a minimum of time.

Big Blue scientists developed a new classification task and found that a quantum algorithm using the quantum nucleus method was able to find relevant features in data for precise labeling, whereas for classical computers, the whole of data sounded like random noise.

“The routine we use is a general method that can in principle be applied to a wide range of problems,” Kristan Temme, researcher at IBM Quantum, told ZDNet. “In our paper, we formally prove that this quantum kernel estimation routine can give rise to learning algorithms which, for specific problems, surpass classical machine learning approaches. '

End-to-end acceleration

To prove the advantage of the quantum method over the classical approach, the researchers created a classification problem for which data can be generated on a classical computer, and showed that no classical algorithm can do better than a random response to answer the problem.

However, when they visualized the data in a quantum feature map, the quantum algorithm was able to predict the labels very accurately and quickly.

'This article can be considered as an important step in the field of quantum machine learning, because it proves an end-to-end acceleration for a quantum kernel method implemented in an error-tolerant manner with realistic assumptions', concludes the research team.

Limited use by current quantum hardware

Of course, the classification task developed by the scientists at IBM was designed specifically to determine whether the quantum nucleus method is advantageous, and is still far from ready to be applied to any kind of business problem at more. large scale.

According to Kristan Temme, this is mainly due to the limited size of IBM's current quantum computers, which so far can only support less than 100 qubits. A far cry from the thousands, if not millions, of qubits that scientists believe are needed to start creating value in quantum technologies.

At this point, we cannot cite a specific use case and say “this will have a direct impact,” adds the researcher. “We have not yet realized the application of a 'big' quantum machine learning algorithm. The scale of such an algorithm is of course directly linked to the development of quantum material. '

Theoretical result which opens the door to other research

IBM's latest experiment also only applies to a specific type of classification problem in machine learning, and does not mean that all machine learning problems will benefit from the use of quantum nuclei.

But the results open the door to further research in this area, to see if other machine learning problems could benefit from the use of this method.

Much of the work therefore remains moot for now, and the IBM team has recognized that any new discovery in this area has many caveats. But until quantum hardware improves, researchers are committed to continuing to demonstrate the value of quantum algorithms, if only from a mathematical standpoint.