You're all just underperforming sentient meat

Will quantum computing really make much of a difference to neural networks? I thought they were way more about parallel computing and limited by how big a GPU farm you can build. Pls forgive ignorance

>>8723022
Clock speed. Quantum computer clock speed is measured in KILOHERTZ. Not gigahertz, not megahertz, but KEEEELOOOHHHHUUUURRRRRRTTTZZ

>>8723022
The most powerful calculation on a quantum computer so far is something like 3x15, but that is great. Imagine calculating 3x15 on 4 atoms.

>>8723413
rip headphones

>>8723417
Underrated post

>>8723413
so were the first normal computers

>>8723022
There's one bug thing you don't mention:
>Not every algorithm is amenable to quantum speed up

>>8723022
Yeah, but aren't we the ones who designed them?

Couldn't we just design something better?

>>8723034
This will be vague, anyway:

Quantum computers are about computing multiple states simultaneously. For example you are looking for primes. With traditional computers you'd start at 2, check whether it's prime by dividing it by all numbers smaller than it, continue with 3 and so forth. With quantum computers you can imagine one state being all numbers at the same time. Then you apply all necessary operations to determine whether a number is prime or not to that state and after that, the state only contains primes. Then you ask the state what it is, and it will give you a prime. You have to do this multiple times to get all primes, which is the disadvantage of quantum computers.

Neural networks are an optimization problem. It's actually very simple. A neural network is basically just a function with a lot of parameters (depending on the complexity of the problem dozens, hundreds or thousands). How exactly those parameters are chosen is up to the problem. Then a loss function is defined for your problem (say, the mean squared error if the problem is a classification) and parameters that minimize that loss function are chosen. If you now imagine all parameters in the neural network being quantum states, you can basically "see" the whole parameter space all at once. That will make learning a lot more efficient.

>>8723847
>Quantum computers are about computing multiple states simultaneously
It's a bit more nuanced than that

>With quantum computers you can imagine one state being all numbers at the same time. Then you apply all necessary operations to determine whether a number is prime or not to that state and after that, the state only contains primes. Then you ask the state what it is, and it will give you a prime. You have to do this multiple times to get all primes, which is the disadvantage of quantum computers.
Are you trying to describe Shor's algorithm? Because it is nothing like that.

File: X on SCI.png (31KB, 694x968px) Image search: [Google]

31KB, 694x968px

>>8723022
>strong AI

>>>/x/

dude

quantum

lmao

>>8723851
I'm not sure what the point of your post is. You are free to add whatever you may know. I suggest you do that.

>>8723847
>>8723851
also

>If you now imagine all parameters in the neural network being quantum states, you can basically "see" the whole parameter space all at once.
If all the parameters are encoded in some quantum superposition then they all "exist" at once, in the sense that you can effectively operate on them all at the same time but you can't "see" them when they're in superposition.

An example of this is the quantum algorithm for solving systems of linear equations. Basically if you have vectors [math]\vec{x}[/math] and [math]\vec{b}[/math] and matrix [math]\hat{A}[/math] such that [math]\hat{A}\vec{x}=\vec{b}[/math] then you can input a quantum state vector [math]|b\rangle[/math] "parallel to" [math]\vec{b}[/math] and receive the similarly defined state [math]|x\rangle[/math] as the output. This is great but you can't access the vector components of your answer without repeating the process [math]O(n)[/math] times since measuring anything about [math]|x\rangle[/math] destroys the state, this kills any speedup it has over classical equivalents.

However, you can still get expectation values or inner products with [math]|x\rangle[/math] which can be useful.

>>8723866
My point with this post is that while some quantum algorithms exist that may have some application in ML, their scope isn't as broad as you might think due to their own inherent limitations.

Another point with this and the previous post is that while quantum computing is somewhat about computing multiple things simultaneously, that's only useful if you can

a) Find a way to get it to spit out the correct answer. This is a clever part of Shor's algorithm where it's achieved by causing correct and incorrect answers to constructively and destructively interfere.

b) Get useful information out of the resulting superposition without actually knowing what it is.

>>8723888
Thanks for elaborating.

>>8723022
Nice word salad OP. Let me sum that up for you:
Accodring to D-wave's founder quantum computers will outperfom humans in every aspect by 2028. Will the stong AI that arises from this exterminate us or become a tool to make our lives easier?

See? You wasted so much effort trying to sound smart, its painfully obvious you need others to view you as intelligent.