Ready to put quantum machine learning into practical applications?

Banking institutions often understand and track customers' transaction behavior when they use their bank cards. For example, if someone goes on vacation in Sweden, he can pay by credit card instead of carrying cash. However, once someone else uses it, the bank that issued the credit card cannot decide to block the transaction. After all, there is no evidence in the card swiping records that someone else misappropriated the card.

Banks’ machine learning algorithms make billions of these decisions every day. This is known as the "average classification problem" in computer science, and these models must decide whether a transaction matches a customer's normal purchasing behavior. For traditional machine learning algorithms, this problem is mainly solved by profiling the consumer's payment history and other interaction information with the bank, which is a computationally intensive and imperfect process.

Are quantum machine learning algorithms at the forefront of technology ready for such practical applications? Industry experts are divided on the answer to this question.

#When will quantum machine learning arrive?

Richard Hopkins, a distinguished engineer at IBM and a member of the Royal Academy of Engineering, explained that another option may be to use quantum machine learning algorithms.

He pointed out that traditional machine learning models require a lot of time and resources to train to identify and weigh all the different characteristics of a transaction to determine whether the transaction is suspicious. In contrast, quantum machine learning models use superpositions of qubits to observe these features simultaneously and therefore have the ability to find answers to very difficult classification problems much faster.

In Hopkins’ view, although the field of quantum machine learning is still in the experimental stage, it may only take 5 years to see quantum machine learning algorithms used in various fields such as fraud detection, drug research and development, and computer vision. Applications.

But not all experts agree with this view. Dr. Maria Schuld, a researcher at quantum start-up Xanadu and co-author of the book "Machine Learning for Quantum Computers", believes that in the long run, quantum machine learning undoubtedly has huge development potential, but realizing practical applications of quantum computing currently seems far away. .

She said, "We are scientists and usually make business cases for the science we study. We do this not because we know it will work, but because we hope there will be some exciting results."

Quantum machine learning is still in the experimental stage

Quantum machine learning is a relatively new field. Although research papers on the subject have been published since the mid-1990s, quantum machine learning has only really begun to attract the attention of the scientific community in the past five or six years.

Schuld introduced that quantum machine learning has two broad research areas. One is to use quantum computers to accelerate traditional machine learning algorithms, such as Gibbs samplers; the other is to use quantum computers as Models, specifically using the quantum chip itself as a basis, train the model in a manner similar to traditional neural networks.

Even so, the field as a whole is still in a highly experimental stage. She explained that while machine learning algorithms could prove to have "quantum advantages" over traditional algorithms in some cases, it would be some time before they could be envisioned for real-world applications.

This has been overshadowed by the enthusiasm for quantum machine learning in multiple studies, often published on the premise that quantum advantage was achieved in a single, narrow use case. "This is interesting from an academic perspective, but it doesn't say much about possible applications of quantum computing," Schuld noted. "After all, many of the problems addressed in these studies were carefully coded to work on a quantum computer." In other words, they're only good at solving very specific problems in very specific ways with a quantum computer."

This isn't a problem with traditional machine learning techniques, many of which have Advantages that can be generalized to more problems. In contrast, quantum machine learning researchers still struggle to adapt a method to a range of tasks. For this reason, Schuld clearly pointed out that it is not yet time to use quantum chips as the basis for new machine learning models.

Quantum machine learning needs to prove itself in reality

Although quantum computing researchers agree that in certain circumstances, quantum machine learning can perform far better than traditional Machine learning, but experts are divided over how soon practical applications of the former will emerge.

Hopkins admits that the field still lacks a common template for mapping problems to algorithms and generating solutions, which is also a problem that traditional machine learning once faced, but once it is done, then leveraging this technology becomes very easy.

“Quantum computers allow users to make better, more accurate decisions based on higher-dimensional data sets. We’ve proven this in theory, we’ve proven it in the lab, and we’re increasingly The closer we get to proving that in reality,” he said.

But are quantum machine learning models ready for this? Schuld thinks not. She said, "Until the underlying hardware improves, we don't know whether there will be quantum machine learning algorithms that can solve a large number of practical problems. Unless we have error-free machines, there will be a limit to what can actually be done using quantum machine learning. As far as practical In terms of applications, it's a bit of a shame."

Hopkins disagrees, but acknowledges that it's unlikely we'll see a quantum computer capable of training a ChatGPT-type model anytime soon. He said: "It is impossible to achieve this on a quantum computer with only 433 qubits, but we are making progress every year to expand the number of possible quantum machine learning experiments. People will gradually see quantum machine learning models change It has to be more versatile.”

The above is the detailed content of Ready to put quantum machine learning into practical applications?. For more information, please follow other related articles on the PHP Chinese website!