Can Quantum Computer be used for Machine Learning?

Quantum computing is advancing rapidly. But how exactly quantum computing fits in the training is hard to tell. The discussions revolved low level to for engineers. I'll try to digest it at as high level as possible.

Quantum Computing

You probably heard of the magical state of quantum physics. It has the state of 0 and 1. And they interact in superposition and entanglement (there're tons of YouTube explaining such). The memory requirement reduces drastically as a result. And the ML with a large variables will be made possible.

Even today, companies like Nvidia and Google are racing to produce bigger more efficient hardware every year.

Quantum will be the game changer (there's no doubt those 2 companies are interested in the frontier of the quantum games as well).

What can you use?

When it comes to machine learning, you will not get away from Google's Tensorflow library. They released Tensorflow Quantum in 2019, which will let you write like a regular python to invoke Google's quantum Cirq.

In this Google's blog post, a student discusses the difference in what you can do with the quantum machine learning over the conventional neural networks (it's ironic neural network has been popular only in the past 10 years and it's already become conventional). The technique is called variational quantum circuits (QVC).

Without going into detail, the important to remember is that QVC takes in quantum data instead of the classic data type like int. Only at the measurement stage, the data is moved to the classical computers where the machine learning magic happens (ie: calculate loss function using Adam and update parameters.) In here, classical computers mean it's GPU.

This physics student ended up using Tensorflow Quantum for reinforcement learning. His testimonials describe it as impossible to code quantum computer without Tensforflow library. The library is already indispensable for all.

Having said, the documentation for tensorflow quantum throughout is full dedicated for researchers. I'm not clear yet if there's any practical application useful in quantum.

Timeline

Google announced in this year's annual announcement: their roadmap for available/useful quantum computation will spin out in the next 10 years. Tensorflow Quantum's purpose is more to test the algorithm for researchers at this moment.

The timeline of quantum computers:

IBM is the big player in the market with their free open-source playing ground of Qubit. They are more optimistic than Google. They plan to offer a thousand qubits for users over the cloud by 2023. That's only 2 years ahead.

Either way, the quantum computers have reached supremacy last year in a narrow domain. Google famously said their noisy 53 qubits did calculation in 200 seconds of something that would have taken 10,000 years on the classic computer. That means even at today's window, there's a narrow domain that quantum computers can excel at.