Another Hardware Alternative for ML and AI: Quantum Computing

By Yina Moe-Lange

 

Quantum computing is continuing to scale up and with the recent announcement from the Vancouver based quantum computing company, D-Wave, of their 2,000-qubit processor it does not show signs of slowing down. D-Wave is the first quantum computing company that has made the technology available for commercial use. The quantum computing processors are in direct competition with the more tradition types of chips used for Machine Learning and AI like GPUs and the newly announced second-generation TPU from Google.

 

The important part of quantum computing is that it replaces the traditional way of thinking of computing. By replacing the conventional bit, 0 or 1, with a new type of information, it opens up to exponential amounts of possibilities. The qubit can be in the superposition state where it is neither +1 or -1 yet, in a sense it is both, and it is this that allows for the superfast computing.

 

The D-Wave quantum computers use the process of annealing. This involves a series of microscopic magnets to be arrange on a grid. Each magnetic field influences each other and then they orient themselves into a position to minimize the amount of energy stored in the entire field. It is during this process, that one can change the strength of the magnetic field from each magnet so that the magnets orient themselves in a way to solve specific problems. To get to the solution, you begin with high amounts of energy so it is easy for the magnets to flip back and forth. Then as you lower the temperature, the magnets reach lower and lower levels of energy until they are frozen into the lowest energy state. Here it is possible to read the orientation of each magnet and find the answer to the problem. One can say that D-Wave’s quantum computer is a kind of analog computer relying on Nature’s algorithms to find the configuration of the lowest energy state.

 

 

This is where we get lucky. This specific class of quantum computing happens to be useful for a subset of optimization computing problems, especially those geared towards Machine Learning. Many Machine Learning problems can be reformulated as energy problems. The D-Wave quantum computers are designed to support problems that need high level reasoning followed by decision making. The quantum computing allows for AI systems to imitate human thought processes much more closely than a classical processor. And while the idea of quantum computing can be hard to understand, its use in Machine Learning is clearly opening up new opportunities.

 

In the impending fight between the GPUs and TPUs, there is a possibility that quantum computing will pass in the outside lane. A key element in D-Wave’s quantum computing is that it isn’t necessarily designed to solve every problem but it is addressing the same need in the processing market that GPUs currently fulfill. Google released a paper in which they find that there is a considerable computational advantage when using the D-Wave quantum computer over a classical processor. In many aspects, a quantum computer can do the same thing a GPU can do, just faster, and these days, time is money.