One of the people mentioned in the article, Svore, works at the QuArC group for MS and has worked on quantum machine learning. (QuArC generally works on how to actually design and use quantum computers.)

For example: https://arxiv.org/pdf/1412.3489.pdf

From my (shallow) understanding, you can get improvements by considering the space of possible states of your quantum system to be the space of weights for your model, and then annealing the quantum system (which is equivalent to optimizing the objective function). The quantum annealing is a (significant) improvement over other optimization methods... or something.

My (admittedly fuzzy) understanding is that while running a DNN is a very parallel problem, training it (back-propagation, in particular) is not and is thought to be susceptible to quantum speedup. Google's work in quantum computing is motivated by the same application.

It entirely depends on the problem being solved. I am not sure what their ideas are with machine learning, but essentially quantum computers help with parallelization with algorithms that can take advantage of superpositions.

I think that for pattern matching (DNN) it's not clear, but for parallel stuff the quantum computer is like the ultimate SIMD machine.

The sample they showed did not look like any of the standard .NET languages, so I wonder if they are going with a standalone language. AFAIK most quantum programming languages are actually embedded into another language (C, Haskell, ML) so if they actually implemented a standalone language, that would be interesting. Not sure what the file extension means.

Uh oh, here comes Quantum Basic.

when you look at the example, there was also a teleport.g.cs file which looks like a standard #C file, but possibly auto-generated(?) What do you think?