Quantum Machine Learning - Computerphile

So today we'll be talking about quantum machine learning. You may recall I had an episode on quantum software engineering and there I talked about the complimentary strength of quantum and classical systems and one example of uh combination of quantum and classical systems are quantum machine learning um

systems or software. Um so they kind of replicate everything that you have seen in machine learning but in the quantum space and I'm going to explain that.

So let's start with the quantum bit. So quantum machine learning is obviously quantum plus machine learning, right?

And it literally is very often a combination of classical machine learning techniques with some quantum bits added to it. Let's start with what quantum is just to recall the basic concepts. So if you look at classical algorithms, classical machine learning, you your data comprises bits, right? So and and

you can think of a bit as a vector that can point either say upwards for one or downwards for zero. Right? So these are two the two values of your data and of course you have strings of very large strings of these bits uh representing your your data and and of course you will have several data points to to to

be represented. Now when we go to quantum you see a very natural extension of this. Rather than having only bits that point upwards or downwards, you can think of a single cubit. So if this were a bit, a cubit would be just not like a bit. But instead of only pointing upwards or downwards, it can point to

anywhere on the surface of the sphere. Right? So this can be any arbitrary cubit Q. And once you look at it, once you measure it, it will fall down to the basis of your measurement say 0ero or one with a probability that is proportional to this angle. So your data can represent now not only 0 or one but

can simultaneously represent any linear combination of 0 and one by pointing at any arbitrary point on on this sphere.

This is called a block sphere. And this is the first principle that gives you some level of say parallelism in codes because you can operate on data that are on a superposition state. So they represent both zero and one a combination of zero and one a linear combination of zero and one in a single

state. So that's that's already quite something. But that's not sufficient to give you quantum advantage. Another important principle is that if you have a number of these cubits all pointing to a place on the block sphere, they can be entangled to each other using certain operations. So we'll talk about those

operations a bit later. And the entanglement of these two cubits, let's call this one Q not, this one Q1, will give you a power of operating on one of them and having non-local effect on the other. So you're simultaneously calculating on Q1 and Q0 by doing a local operation on one of them. And also

once you measure them they will fall onto the basis of the measurement um uh of of the basis that you have chosen and it could be a partial basis there there is a discussion about that that we can discuss separately. So you will have entanglement which gives you another source of say quantum