Classical machine learning (ML) algorithms have proven to be powerful tools for a wide range of tasks, including image and speech recognition, natural language processing (NLP), and predictive modeling. However, classical algorithms are limited by the constraints of classical computing and can struggle to process large and complex datasets or achieve high levels of accuracy and precision.

Enter quantum machine learning (QML).

QML combines the power of quantum computing with the predictive capabilities of ML to overcome the limitations of classical algorithms and offer performance improvements. In their paper “On the role of entanglement in quantum-computational speed-up,” Richard Jozsa and Neil Linden, of the University of Bristol in the UK, write that “QML algorithms hold the promise of providing exponential speed-ups over their classical counterparts for certain tasks, such as data classification, feature selection, and cluster analysis. In particular, the use of quantum algorithms for supervised and unsupervised learning has the potential to revolutionize machine learning and artificial intelligence.”

QML versus classical machine learning

Zohra Ladha, senior director, of data science and AI at Credence, says QML differs from traditional machine learning in several key ways:

EVENT

Intelligent Security Summit On-Demand

Learn the critical role of AI & ML in cybersecurity and industry-specific case studies. Watch on-demand sessions today.

Watch Here

 

 

  1. Quantum parallelism: Quantum algorithms can take advantage of the unique property of quantum systems known as quantum parallelism, which allows them to perform multiple calculations simultaneously. When processing large quantities of data, such as images or speech, can significantly reduce the time needed to solve a problem.
  2. Quantum superposition: Quantum superposition allows a quantum algorithm to represent multiple states simultaneously. This can enable it to explore possible solutions to a problem, leading to more accurate and efficient solutions.
  3. Quantum entanglement: Quantum algorithms can also use the property of quantum entanglement, which allows quantum systems to be correlated in ways that classical physics cannot explain. This can enable quantum algorithms to perform certain tasks more efficiently than classical algorithms.

 

Traditional machine learning algorithms, which rely on classical computing techniques and lack these quantum capabilities, may be slower or less accurate in certain cases.

The QML journey: From research to the real world

Research on quantum machine learning began in the 1980s. In the late 1990s and early 2000s, researchers developed quantum neural networks to demonstrate the potential of quantum systems for machine learning that can be trained to recognize patterns in data. These networks have since been applied to a wide range of real-world problems.

A decade later, researchers developed quantum algorithms and software tools for machine learning tasks. These included quantum versions of popular machine learning algorithms such as support vector machines, decision trees, and neural networks.

The development of quantum computers has also been a key factor in the growth of QML. In the 2010s and 2020s, several companies and research groups developed quantum computers that could perform machine learning tasks. These included both gate-based quantum computers and quantum annealers. By the 2020s, QML began to be widely adopted in applications including pattern recognition, natural language processing, and optimization.

Today, one of the most promising applications of QML is drug discovery. Traditional drug discovery processes can be slow, expensive, and inconsistent. QML has the potential to accelerate the process. “After our initial success in finding a COVID-19 therapeutic molecule, we wanted to extend the space to now generate smaller molecules,” said Nikhil Malhotra, global head, of Makers Lab at Tech Mahindra. “Quantum GAN or hybrid GAN generation is something that we are trying for small molecules. This, I believe, would advance drug discovery and even novel drug creation significantly.”

Financial markets are another area where QML has shown promise. A 2021 paper from JPMorgan’s Future Lab for Applied Research and Engineering concluded that QML can perform tasks such as asset pricing, predicting volatility, predicting the outcome of exotic options, fraud detection, stock selection, hedge fund selection, algorithmic trading, market-making, financial forecasting, accounting and auditing, and risk assessment much faster and more accurately than classical algorithms.

The 2023 promise of quantum machine learning

“QML is an exciting and rapidly developing field that has the potential to significantly impact a wide range of industries and applications,” said Ladha. For 2023, she predicts that quantum algorithms have the potential to perform certain machine learning tasks faster and more accurately, especially for tasks such as image and speech recognition, which require processing large amounts of data. She also points out that QML can address optimization problems that often arise in machine learning tasks and are difficult to solve using classical algorithms. Ladha predicts that quantum algorithms’ ability to solve these problems more efficiently could benefit finance and logistics.

Cybersecurity is another area where she predicts QML to make an impact. “By developing more sophisticated algorithms to detect and prevent cyberattacks, quantum machine learning could improve the security of systems,” she said.

Digging a little deeper into the technology itself, Malhotra said that he expects to see a vast majority of ML algorithms, particularly those on artificial neural networks, to be tried on the quantum machine as quantum machine learning algorithms. “We have seen early renditions like QNLP, Q-GAN and even reinforced learning on quantum circuits. I anticipate the trend to grow in 2023,” he said.

Quantum machine learning challenges remain

QML is a big deal because of its promise. Evidence suggests that machine learning models can be trained for higher accuracy with fewer data than they can with our current classical techniques. However, according to Scott Buchholz, global quantum lead and CTO, of government and public services, at Deloitte Consulting LLP, the answer to the questions, “How much less?” and “How much faster?” changes regularly due to the following challenges QML:

 

  • “In terms of hardware, the most powerful quantum computers in existence today are still limited — particularly when compared to today’s most powerful servers. We anticipate that changing in the coming years because there is far more room for quantum computing technology to advance and grow.
  • “In terms of software and algorithms, quantum computers work fundamentally differently than today’s computers. As a result, researchers are trying to figure out the best ways to map problems onto quantum computers (and indeed, determining which problems might be useful to solve on quantum computers). As we come up with better-generalized mappings, it becomes easier for others to ‘bring their problems’ to quantum computers.

“For years, QML has been — and continues to be — an area of active research. As hardware and software maturity improves, we are likely to see organizations start evaluating the use of QML in production workloads,” Buchholz went on. “Since we are still some years off from having a machine that could run production QML workloads, we are continuing to advance the state of the art in QML as the hardware continues to improve. But I anticipate incremental QML progress throughout 2023 — that is, continuing to improve the techniques to scale volume, load data, and run models.”

Translate »