• 3 MIN

Quantum Computing: A powerful tool for potentially enhancing machine learning

Quantum Computing: A powerful tool for potentially enhancing machine learning

By SMU City Perspectives team

“The new technology of quantum computing used in machine learning (QML) is very exciting and a first in-depth exploration of the use of hybrid QML in credit scoring shows great promise.” - Associate Professor of Information Systems (Practice) Paul Robert Griffin


In Brief

  1. Quantum computers offer huge potential in solving complex problems but the technology is still in its infancy stage. While it might take years for quantum hardware to be of higher quality and become more accessible, the technology can still be explored effectively today.

  2. Research suggests that using a hybrid classical/quantum model improves the efficiency of the machine learning process and could even lead to better decision-making in industries such as credit scoring.

  3. Classical computing will continue to play an integral role in solving grand challenges like climate change, and all individuals and organisations need to do their part to create solutions that quantum computers can then take further.

At a panel discussion titled ‘The Brave New Quantum Economy’ at the 2022 World Economic Forum (WEF) Annual meeting in Davos, panellist and physicist Jeremy O’Brien shared his opinion that quantum computing is “destined to be the most profoundly world changing technology humans have yet uncovered”, even more significant than the launch of artificial intelligence (AI). Due to its unique features of entanglement and superposition, he and other experts in the field anticipate that this powerful tool holds the key to overcoming the climate crisis and unlocking unparalleled developments in healthcare and energy. The catch is, we first need to get the technology right. 

Associate Professor of Information Systems (Practice) Paul Robert Griffin, says that quantum computers can “find solutions to problems we haven’t yet thought about” and offers huge potential to solve existing ones that currently seem computationally unsolvable (e.g., the travelling salesman problem). While steps are being taken to build the hardware and software needed to handle these types of complex challenges, focus is also being given to the development of new algorithms that make current, though limited, quantum computers applicable to real world problems. Prof Griffin is especially interested in the application of this technology to machine learning (ML), and he seeks to understand how it can disrupt industries like the financial sector, which currently leads in ML adoption.  


Improving credit scoring model efficiency

In a recent research paper, Prof Griffin sought to understand the benefits of incorporating quantum machine learning (QML) in the credit scoring process. He and his team proposed a novel hybrid classical/quantum model and used a dataset of nearly 2300 small-medium enterprises (SMEs) in Singapore to determine if this hybrid approach could improve the efficiency in model training and eventually, the accuracy in the credit scoring process. According to Prof Griffin, “currently even the best credit scoring models only reject 90% of companies that become insolvent, while also rejecting 15% of companies that remain healthy. Even a small improvement in credit scoring model efficiency can translate to significantly lower rejections for good SMEs and lower risk for those funding the loans, giving potential gains of around US$40 million to regional SMEs.” 

After two years of model simulations, he concluded that “the use of hybrid quantum/classical models showed great promise, given the ease of obtaining comparable results to a purely classical counterpart, and with much fewer epochs for training the neural network model”. The training times for QML are ten times faster than for a classical counterpart, a significant advantage since training time is especially important for large datasets and high-velocity data. While just the tip of the iceberg, the experiment demonstrates some of the benefits QML can potentially bring to industries like the financial sector that utilise ML to improve decision-making.  


Early-stage limitations of QML

With quantum computing being in its infancy stage, there are many limitations to what can currently be achieved. Unlike classical computers which use transistors in silicon chips for bits, quantum computers engineer atomic nuclei, electrons or photons to create quantum bits (or qubits). Due to the complexity of the technology and the sensitivity of quantum particles however, present day quantum processors have limited capabilities and have only just surpassed the 100-qubit barrier. Prof Griffin shares that for his experiment, only 20 qubits were used but to get near the levels of classical counterparts, up to 400 qubits would have been needed together with more complex quantum circuits.

While it might take years for quantum hardware to be of higher quality and become more accessible, the good news is that there are breakthroughs in the technology every week. Many vendors publish roadmaps to show their expected improvements, with IBM, for example, targeting a 4,158-qubit processor by 2025. Software vendors such as Xanadu, Horizon and Classiq are also coming in with great products that allow programmers to build and develop quantum applications with relative ease, and without any training in the field.


Will QML replace classical ML?

Prof Griffin shares that in machine learning, the most effective model depends on the data and use case. In his view, QML is just another, albeit special, type of machine learning that can be used and as such it is unlikely to replace classical ML for all use cases. Nevertheless, he encourages everyone in the machine learning community to experiment with QML for their own problems, especially since platforms such as Pennylane allow developers to code QML models in the same way they code classical ML models.



A future built on collaboration

At the WEF panel, Jeremy O’Brien warned against the mindset that quantum computing will “ride to our rescue and solve climate change”. He encourages individuals and organisations to do everything in their power to tackle the problem using the tools at their disposal, so that when quantum computers become more powerful, these solutions can be taken further. Prof Griffin echoes this sentiment and says that classical computing will continue to play an integral role in achieving society’s grand visions. As demonstrated by his experiment on credit scoring, there is much to gain from taking a hybrid classical/quantum approach and even greater benefits can be unlocked as both types of technology develop further. He envisions a quantum state in the future, where a quantum internet with sensors, networks and computing all work together and lead to new collaborative techniques.