Original article was published on Artificial Intelligence on Medium
Quantum Machine Learning — The Whats, the Hows and the Whys
A sneak peak into the world of Quantum Machine Learning and why it is relevant
Quantum Machine Learning is quickly becoming the new buzzword of the AI industry. With greater understanding and developments in the field of Quantum Computing, it was only a matter of time before its already massive application field encompassed Machine Learning as well. For the uninitiated, the last two sentences must have seemed like a whir of incomprehensible words aboard the ever-moving train of AI, but worry not!
This article will expunge your questions like India will expunge the corona-virus (so keep your hopes up!)
Umm… What is Quantum Computing again?
An ordinary computer chip is made of bits, which can take two states 1 or 0. However, the world is a lot more uncertain than just an on and off state, especially when you go down to the microscopic level. This caused even the best of the supercomputers to perform undesirably. So, Quantum Computing chucks the old ‘1’s and ‘0’s out the window and replaces them with something called qubits. Qubits don’t need to be just on or off, they can also lie on a spectrum between the two. The other thing that qubits can do is called entanglement, meaning two particles are not independent of each other.
This implies that in the realm of quantum computing, you can move information around, even if it contains uncertainty. And once you string together multiple qubits, you can tackle problems that would take even the best computers millions of years to solve.
Why Quantum Machine Learning?
Quantum computers are used primarily in stimulating large, uncertain complicated systems. Therefore, it goes without saying that Quantum Computers are extremely useful in Artificial Intelligence. How Quantum Machine Learning works is by making use of the fact that a qubit can store all 4 binary configurations simultaneously. This means n qubits can store all 2^n binary combinations. So, as you add a qubit to a quantum computer, the computation power of the computer rises exponentially.
Anyone who’s worked with Machine Learning models before would know how painstakingly slow they can be. Quantum Machine Learning makes this computation faster using its hardworking elves (the qubits) to increase efficiency. A number of quantum algorithms have been proposed for various machine learning models such as neural networks, support vector machines, and graphical models, some of which claim runtimes that under certain conditions grow only logarithmic with the size of the input space and/or dataset compared to conventional methods. QML algorithms are often based on well-known quantum subroutines (such as quantum phase estimation or Grover search) or exploit fast annealing techniques through quantum tunneling and can make use of an exponentially compact representation of data through the probabilistic description of quantum systems.
Long story short, Quantum Machine Learning is fast and efficient. And by bucket-loads.
Quantum Machine Learning sounds great — I’m going to use it EVERYWHERE!
Not so fast, José. Don’t get me wrong, I’m all for QML, but there’s still some time before quantum computing could be used to do any significant machine learning. We are talking years here, and many of them. This is because a lot of extra qubits will be needed, i.e. we are qubit limited. Also, apart from faster computation, which is also not always guaranteed, it is not entirely clear whether QML will reap any benefits over its classical counterpart. Though, signs of limitations in the current software algorithms are emerging, and it is clear that a new hardware will soon be required.
Until then, to be or not to be, that is the QML question.
Some last qubits of information
As and when (or if, for the cynics out there) QML becomes a reality, it will see widespread applications in all walks of life. Artificial intelligence will witness rapid acceleration and growth. Google is already using them to improve the software of self-driving cars. They’ll also be vital for modelling chemical reactions, and where a rapid classification needs to be made, typically in military applications.
QML has been theorized for decades, but once it becomes a widespread reality, the possibilities it will bring are endless.
References and further reading