Evrim Yazgin
Cosmos science journalist
“When we talk about quantum computing, a lot of the time we focus on quantum hardware. My message is that quantum software and quantum algorithms are equally important.”
This is the message of Dr Muhammad Usman, head of quantum systems at the Data61 – an arm of Australia’s national science agency CSIRO focusing on everything related to data science, including artificial intelligence, cyber security, large data processing, and software design and engineering.
“You might be wondering, what is ‘61’?” Usman says. “61 is Australia’s national telephone code. CSIRO has a global outlook. We are not only in Australia. But having that code 61 maintains the national identity.”
Data61 was established in 2016 and now has more than 1,000 staff, including about 300 PhD students from 70 countries. It is partnered with more than 30 universities globally.
Usman is a quantum computing specialist who completed his PhD at Purdue University in the US in 2010.
“When I came to Australia, I started working at University of Melbourne where I was an associate professor,” he says. “I was working on developing quantum computers, asking questions like: how can you make single qubits? How can you make 2-qubit gates? And then how can you have millions of qubits working in a quantum processor?”
After moving to CSIRO about 2 years ago, Usman brought together a team at Data61 to develop quantum algorithms and software.
Cart before horse?
There’s no shortage of interest in quantum algorithms.
“We have got a project with the Queensland government where we will develop quantum algorithms for 2032 Brisbane Olympics. We will optimise traffic and pedestrian workflows athlete movement for the Olympics,” Usman explains.
But hang on. There currently aren’t quantum computers which are operating on the scale that need special software. Software requires hardware to run on, doesn’t it?
“I think it’s a fair question,” Usman says. “When I started working in the field of quantum computing about 10 to 15 years ago, this question was even more important. Quantum computers are one thing, but there wasn’t even a single qubit at that time.”
“In the last few years, there has been remarkable progress on quantum processor and hardware development,” Usman says. “We see large companies such as IBM, Google, Microsoft internationally and some companies in Australia all developing larger quantum computers. The current status is we have about 100 qubits. They are noisy. By noisy I mean they have errors. So they are not perfect. They don’t do the tasks that we want them to do, but they are improving gradually.”
Error correction is one area in which software and algorithms can help the development of hardware.
“There is this back and forth,” Usman explains. “We are actively working on error correction schemes. I have 4 PhD students working on quantum error correction, and they continuously interact with the current state of processors to tailor algorithms which can improve hardware and improve the software as well.”
Noisiness in quantum systems comes from the same quantum mechanical effects that makes such devices so promising for the development of computing speed and power.
Quantum computing basics
Conventional, or “classical”, computers store and process information in the form of 0s and 1s – each binary value is called a “bit”.
Quantum computers are not binary. Quantum bits, or “qubits”, can have values of 0, 1 or a mixture of 0 and 1. This is due to a quantum mechanical property known as superposition where a quantum object isn’t in one or another physical state, but has a probability associated with those states.
“When we try to process information using electron spin, photon polarisations, etc., we have access to those special quantum mechanical properties. For example, electron spins can be placed in a superposition state. If up-spin is 0, down-spin is 1, then a superposition of up-spin and down-spin will be mixing 0 and 1. That makes huge difference when doing computing, because now you have a large computational space which you would not have in classical system.
“The whole idea in quantum software design and quantum algorithm is to leverage that very large computational space.”
Usman’s team is working on several quantum software projects including more efficient artificial intelligence (AI), traffic optimisation and even improving research into climate science and emissions reduction schemes.
Quantum artificial intelligence
“Conventional artificial intelligence systems are very efficient. For example, they can easily identify people or objects in images, once they are properly trained,” Usman says.
“However, artificial intelligence is also very easy to trick or fool. You can very easily change some of the pixels in an image or video, and the same artificial intelligence system will give you wrong answers. That is extremely important for security sensitive applications.”
Quantum AI, on the other hand, is very robust. This isn’t just about how powerful quantum computers are – it’s about the quantum nature of the artificial neural networks themselves.
“In classical deep neural networks, you have several layers of neurons. You feed pixels of an image or a video, and then, based on that input, all the neuron weights are changed. They update gradually and the neurons learn the features and make decisions of their own.
“But the important thing is that in classical machine learning, all the neurons are independent. They are not interacting with each other. They are connected to each other, but they are trained independently.
“Neurons in quantum artificial intelligence systems are entangled.”
Entanglement is a quantum mechanical property where entangled objects can communicate instantaneously no matter how much they are separated. The 2022 Nobel Prize in physics was awarded to scientists who first showed this mysterious phenomenon in experiments.
“So, they are very strongly entangled systems. When you feed information in quantum neural networks, all the neurons collectively make decisions. They change their weights collectively rather than individually, and that is what creates the robustness.”
Preparing for a quantum future
But the question still remains: isn’t all of this a little premature given there aren’t quantum computers to run the software?
“Quantum software and algorithms are not easy to develop,” Usman explains. “They take lot of time and benchmarking and understanding and innovation to create. The idea is that, when quantum computers exist in the next few years, software and algorithms won’t magically appear. It has to be developed along with the hardware.
“If you look at road maps, a large-scale quantum computer is anticipated to exist in maybe 10 years time. So that 10 years is needed to develop software and algorithms as well. That’s why our team is focusing on developing these algorithms and software platforms so that when the quantum computer is ready to solve problems, we have the solutions ready to deploy.
“Otherwise, there will be hardware sitting there, but there’ll be no software to run on it.”
Usman is optimistic that quantum computers that do things that no classical computer can do will emerge very soon.
“I always say that the science is very clear. We know that quantum processors can be developed and they can work. The only problem is the engineering issues – because these are highly complex systems – like making qubits, shielding them from environment, designing controls which are very, very accurate and can do lot of qubits on a large scale with coherence.”
But he stresses the importance of quantum software.
“Even if you have a large-scale quantum computer, it cannot do anything useful unless you have an algorithm or software which can run on top of it. Imagine you have a laptop, but you don’t have an operating system. It’s just a dead piece of hardware. You really need to have the software engineering, the algorithms which enable quantum processors to do useful things.”
