NB: If you are a highly technical PhD type software person, best to skip this and go straight to Dr Javier Mancilla Montero, PhDpost hereand then readthis paper.- my dramatic oversimplification may only serve to annoy you.
For the rest of humanity lets take a look at Quantum Machine learning in simple non technical terms.
Imagine in your minds eye that you’re solving a jigsaw puzzle of a famous painting.
A big jigsaw - lets say 100,000 pieces.
Normally, you’d try to collect and fit together every single piece to complete the picture.
That is sort of how machine learning works, it takes in all of the data “pieces” and puts them together based on its knowledge of the picture on the front of the box.
You feed AI lots and lots of data, it “machine” learns which jigsaw pieces create certain images.
Show it hundreds of thousands of images of the front of Jigsaw boxes, then feed it lots of pieces and it will eventually work out what jigsaw pieces are most like those on the front of the box. ( I told you it was highly over simplified).
But what if you could pick just a few key pieces—the ones that best represent the entire scene—and still work out exactly what the painting was? (Unlike traditional machine learning) you just have to give your AI the important bits of your jigsaw.
You give it a framework, for example the “Hey Wain” jigsaw is a mill next to a river, with a man driving a cart up the river, and then feed it a set of the most important pieces. Some corner blocks, some centre blocks and a few in the middle.
That’s similar to what Dr Javier Mancilla Montero’s experiment did with machine learning using quantum techniques when working on credit data.
What He Did
Selective Data Use: Instead of feeding the machine learning model thousands of data points (like having every puzzle piece), he chose only 10 crucial samples (5 from each class). These were the standout pieces needed to form the picture puzzle.
Quantum Optimisation: The method, called QUBO-SVM, works like having a smart helper who knows exactly which puzzle pieces are the most important. Using a simulation tool called Pulser, the system mimicked quantum behaviour to pick out these key pieces quickly and efficiently - he ran also ran the test on a live quantum computer.
Superior Performance: Even with just these few pieces, the model consistently achieved a high accuracy score (AUC above 0.70) on a test set, outperforming traditional methods of machine learning that uses thousands of data points by feeding it just a few pieces from each class. The experiment was a result of this paper.
Why It Matters
More efficient: Just as finding the right puzzle pieces can save time and effort, selecting only the most critical data points reduces the need for massive amounts of data. This may make the machine learning process faster, less expensive, and less resource-intensive.
New Possibilities for Small Datasets: In many real-world situations, gathering huge amounts of data isn’t practical. Think of rare medical conditions or niche markets where data is limited. This approach shows that even with scarce data, you can still get reliable and accurate insights.
**Imbalanced Data -**In many IT scenarios, banking, weather prediction, cyber-security, ITOps you don’t get balanced datasets, they change all of the time, this approach may only need snap shots of datasets. This is a big problem for example in Cybersecurity because the patterns of attack vectors, methods of hacking and systems used change constantly as people try to hack into your systems. There is no one standard model, data changes.
Pioneering a New Approach: This work challenges the old idea that “more data is always better.” Instead, it suggests that smarter data selection—using innovative quantum methods—can lead to better results, much like assembling a complete puzzle with just a few key pieces.
Potential Massive QML Impacts and Down-the-Line Benefits
QML is new - very new, and there are a distinct lack of resources currently, but it is a promising new technology available now, it is currently at its earliest stages, and it is almost certainly a game changer for AI.
Cost Savings and Efficiency: Businesses and researchers may achieve strong results without investing heavily in data collection and storage. This means lower costs and faster decision-making.
Access to High-Quality Insights: Industries with limited data can benefit greatly. For example, healthcare, finance, and specialised manufacturing might use these methods to gain accurate predictions and insights even when data is hard to come by.
Innovation in Machine Learning: These breakthroughs pave the way for new techniques that could redefine how we build models in the future. With quantum optimisation, we might see more agile and responsive systems that work well with less data and use less electricity.
Advancement in Quantum Computing: As quantum technologies evolve, integrating them with machine learning could lead to more robust models that not only save time and money but also open up new possibilities in fields that depend on quick, accurate data analysis.
How It Works
Quantum machine learning works by harnessing the unique abilities of quantum computers to process data in a way that’s more like seeing the whole puzzle at once rather than piecing it together slowly.
And this is where it’s going to start getting a bit freaky…
Instead of examining every single data point one by one, it uses quantum principles like superposition (being in many states at the same time) to quickly sift through all of the possibilities and find the most important information.
This means that even with just a few key pieces of data, a quantum machine learning model can spot patterns and make accurate predictions much faster and more efficiently than traditional methods.
Why It Works
The multiple possibilities come from the very nature of quantum bits, or qubits, which can exist in several states at the same time—a concept known as superposition. In traditional computing, a bit is either 0 or 1, but a qubit can be all possibilities all at once until it’s measured (again a massive over simplification). This is where simple explanations get harder.
If you imagine a maze with a mouse running around it trying to find its way out, the mouse must try each individual route until it gets through the maze. With Quantum computers its like computing an instant map of the maze where all exit possibilities are computed at the same time.
This means that a qubit naturally represents many potential outcomes simultaneously, with each outcome having a certain probability determined by its quantum state.
These probabilities are mathematically described by something called a wave function, which is a core part of quantum mechanics. So yes, we do understand where these possibilities come from: they are a built-in feature of the quantum world, governed by well-established physical laws.
Why they are there? - Scientists are still debating, this is where the phase “if you say you understand Quantum Computing, you don’t really understand Quantum Computing” comes from. There are debates, positions, ideas and bundles of papers - but the reality is we are still working it out and it raises some quite deep philosophical questions.
But the important thing to know is that it works. At least for credit data ;-)
In a Nutshell
The experiment is like solving a jigsaw puzzle by finding just the right pieces to reveal the whole picture. By smartly choosing a few key data points using quantum optimisation, its possible to build a model that outperformed traditional methods that need tonnes of data for accuracy. This innovative approach could transform industries, reduce costs, and inspire a new era of efficient, effective machine learning.
The other thing to note is that this technology is delivering these results at an early and nascent stage of development - the technology is currently the worst it will ever be and its only going to improve.
It may well be a core component of achieving super intelligence, and is already resolving problems that were otherwise unsolvable in banking analysis, pharmaceuticals, transport, logistics, healthcare, IT and just about every other industry you can think of.
So if you are currently thinking about investing in AI - make sure you have a smart Quantum Machine Learning person on your team - and if you are in AI or working with AI pick up some QML skills quickly.
There are Quantum Companies working in #banking #chemistry #defence that can help you work out how best to use this technology to gain competitive advantage today.
Companies that can help you
Falcondale provide QML for Banking companies (EU)
SquareGen provide QML for credit teams (EU)
Quantica Computacao provide QML services (India)
Quantinuum have a QML team (EU/USA)
Qulabs Software India Pvt Ltd provide QML solutions (India)
JoS QUANTUM provide QML solutions (EU)
PennyLane provide QML software (Intl)
QuEra Computing Inc. Provide QML software (Intl)
Terra Quantum AG Provide QML solutions (EU)
Google Circ - provide QML software (Intl)
Sources for Further Reading:
- Research on QUBO-SVM (Quantum Unconstrained Binary Optimization for Support Vector Machines)
- Quantum computing simulation tools like Pulser
- Comparative studies on machine learning models such as XGBoost and Logistic Regression
This breakthrough shows that sometimes, focusing on quality over quantity isn’t just smart—it could be the future of technology and data analysis.
