Comment by asdfman123 | Hacker News Reader

Comment by asdfman123 | Hacker News Reader

Parent Full thread

I'm not sure about the next 20 years, but quantum computers are really good at running calculations in parallel and finding a result. I'm not 100% sure I'm accurate, but I believe if an algorithm is a good candidate to run on, say, a Spark cluster, a quantum computer would be much better at it.

I think the biggest change will be that quantum computing can really kill at machine learning, upping the numbers of nodes in neural networks with an exponential relationship to the number of qubits.

It might turn neural nets into a neat toy that Facebook uses to show us sort-of good ad predictions to something that really cuts into a wide range of things that only humans are capable of doing.

Quantum computers are not generically good at all parallel problems. They have better asymptotic runtime for “brute force search” type problems, where you know there is some answer but you have to guess-and-check repeatedly. Roughly they change a 2^n brute force algorithm to a 2^(n/2) algorithm in that case. The main area these problems are important is cryptography.

So what you're saying is I should short sell Bitcoin?

You can read more about this issue here: https://en.bitcoin.it/wiki/Quantum_computing_and_Bitcoin

The article says "Attacking Bitcoin keys would require around 1500 qubits". So with 28 of these, Bitcoin is compromised?

No, you need 1500 coherent qubits, in a single system.

Would it be possible to evalute n randomly generated neural nets to find the best one?

Maybe in theory, but you'd at least have to fit the entire neural net into the width of the computer. So we'd need a maching with millions of qubits? That's a bit hard to fathom.

> if an algorithm is a good candidate to run on, say, a Spark cluster, a quantum computer would be much better at it.

This is not true. Some parallelizable tasks will benefit from massive speed ups but most won't.

Reply on news.ycombinator.com