2
$\begingroup$

I'm making my way through several neural network examples and libraries. Landing on this nugget, in creating a basic NN for character recognition.

In calculating weights and biases, looking at the graph below, looks oddly familiar in one of the quantum-related problems to solve - minimizing cost functions and finding local minimums.

Is there an equivalent quantum NN for this and are there examples I can begin working with?

My interest here is in how a parallel quantum system can be more efficient with training data over its non-quantum counterpart.

3Blue1Brown Series

$\endgroup$
  • $\begingroup$ Hi, welcome to QC StackExchange. Regarding My interest here is in how a parallel quantum system can be more efficient, please note that quantum computers are by design not parallel ones. I know that often this notion is used in laymen literature to explain why are quantum computer faster than classical counterparts but it is basically wrong. There are several tasks where quatum computers and not faster than classical ones, some quantum algorithm brings quadratic and some exponential speed-up. $\endgroup$ – Martin Vesely Jan 9 at 19:20
  • $\begingroup$ More about quantum computers speed up is here: quantumcomputing.stackexchange.com/questions/9253/…. Still, thanks for interesting question, I am courious as well. $\endgroup$ – Martin Vesely Jan 9 at 19:22
  • $\begingroup$ @MartinVesely - Thank you for that. With respect to finding which weights and biases minimize cost functions, isn't a quantum solution more efficient? That's how I'm viewing the NN topics I'm looking at, as an optimization landscape. $\endgroup$ – ElHaix Jan 9 at 21:13
  • $\begingroup$ You may be right, I am also curious if somebody post answer on your question. I just wanted bring to your attention that quantum computing does not provide exponential speed-up typical for parallelism always. Regarding optimization, maybe quantum annealers are more suitable for your task, try to type "D-Wave" to seach bar on this website (or Google it). $\endgroup$ – Martin Vesely Jan 9 at 22:29
1
$\begingroup$

I don't think a hello world really exists here. You can have different points of view or goals here. I will give references.

The first one is speeding up parts of the algorithm with a quantum version (here is an example reference). But here, we assume a perfect hardware.

Another one is to apply it to quantum many-body systems. The interesting point here is to have less parameters/weights to work on your problem.

The last one is to apply with quantum circuits that could be run on near-term devices. One way to do so would be to process the data amenable to a realistic-size on a quantum computer, using transfer leaning for instance, and apply a parameterized quantum circuit where its input correspond to the output of a classical NN. For this one, there exists an implementation here.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.