4 ms·

As someone who did most of IBMs quantum course, the difficulty didn't come with the generating code part. It came with the application. How do I use this to spe

As someone who did most of IBMs quantum course, the difficulty didn't come with the generating code part. It came with the application. How do I use this to speed up database access? How do I use this to better commute graphics? Can you make a neural net better with this? Giving people useful applications will generate more interest than you will need.

dr_dshiv 4y ago
Useful applications don’t exist. They surely will, at some point, but not yet. That’s why they aren’t provided as examples.
- jtc42 4y ago
  There are/seem to be very specific scientific applications, mostly around simulation of other systems whose hamiltonians map nicely to that of the computer. For almost everyone on the planet that isn’t very useful, but for that small handful it can be a game changer. … doesn’t contribute much here, but worth knowing! It’s exciting either way.
- amelius 4y ago
  Protein folding (computational chemistry), and breaking everyone's crypto.
  - nabla9 4y ago
    > breaking everyone's crypto Symmetric encryption and hash functions with 256 bits are already quantum proof. Post-quantum algorithms are here in few years. Some of them are already in use in critical applications.
    - t_mann 4y ago
      Importantly, Bitcoin and Ethereum are (inadvertently) almost quantum-proof (because the addresses are hashes of the public keys), just not quite (because the public keys still get exposed during transactions). But at least the latter should be fixable (because the community is quite used to changes to the protocol anyway).
- go_elmo 4y ago
  I have to object, while not many algorithms exists, one of the most famous ones, Grovers algorithm [0] is also one of the most general ones and quite powerful. It reduces search of 2^n items to about 2^(n/2) while using n q-bits. Thats just the square root of a classical algorithm. [0] https://en.wikipedia.org/wiki/Grover%27s_algorithm https://en.wikipedia.org/wiki/Grover%27s_algorithm
- latenightcoding 4y ago
  of course they exist, we just don't have QCs with enough logical qubits yet.
nabla9 4y ago
> How do I use this to speed up database access? How do I use this to better commute graphics? Can you make a neural net better with this? It seems that you have missed some of the basics of quantum computing. For quantum computing to work the system must be isolated from the environment for the period of the computation. You can't use it for anything interactive like graphics. Any practical problem worth solving with quantum computer needs something like years or millions of years to solve using classical computer. Do you have database index where search takes years or graphics problem where output takes years? Also consider the amount of qbits needed. For database search you need quantum computer with same order of magnitude qbits as the database index. Practical problems for quantum computing are mostly scientific problems like simulating chemistry.
keithalewis 4y ago
Factoring large integers in order to decrypt RSA. Quantum computations do not produce numbers, only probability distributions. This is fine if all you are trying to do is factor integers since the answer can be easily verified. This is why massive amounts of money is being thrown at quantum computing.
packetlost 4y ago
QPUs are effectively (and theoretically) high-speed, specialized coprocessors at best. There's no way they're going to be doing anything you mentioned any time soon because that's not what they're good at or built for.
- zmgsabst 4y ago
  There’s some research on AI, eg at Microsoft Research. https://arxiv.org/abs/1412.3489 https://arxiv.org/abs/1412.3489
t_mann 4y ago
You're asking whether they can do things that classical computers can do reasonably well already marginally better, while their whole point is that they (one day, hopefully) can do things that current computers can't do at all (at scale, at least). Most of the speculative applications I've heard are in science, cryptography and AI/ML.
HWR_14 4y ago
IBM has (had?) a quantum course? Was it a free online course? And if one wants to take it for curiosity and not expecting to use the skills (immediately although maybe in the future) would you recommend it?
go_elmo 4y ago
I have to object, while not many algorithms exists, one of the most famous ones, Grovers algorithm [0] is also one of the most general ones and quite powerful. It reduces search of 2^n items to about 2^(n/2) while using n q-bits. Thats just the square root of a classical algorithm. [0] https://en.wikipedia.org/wiki/Grover%27s_algorithm https://en.wikipedia.org/wiki/Grover%27s_algorithm
zackmorris 4y ago
Ya looking at quantum computing (QC) from a programmer's perspective, I see issues with abstraction at multiple levels. Basically what's going on is that it's too narrow (tightly coupled to its implementation details), but also too broad (generalized over many disciplines) to easily transform it into simpler domains that are easier for humans to reason about. So for example, in computer science we can explain basic data structures like binary trees and hash tables from whichever context a student understands. We can use pointers and references from C-style imperative languages. We can use car and cdr from Lisp-style functional languages. We can even jump up to higher levels of abstraction with associative arrays in Javascript. Or make the whole thing in key-value stores like Redis or even a SQL database or x86 assembly if that floats our boat. But with QC, even its "assembly language" of Hilbert spaces and tensors is already so difficult to understand, much less apply, that the terminology feels like gatekeeping. There's no way to easily identify the smoking gun(s) which separate a QC simulation from qubits in the real world. It's like asking an assembly language teacher how a full adder works, but they throw up their hands and say that first we must understand Fermi levels and electron diffusion across a PN junction. Which while important, doesn't answer the student's question. What's needed are simple transforms to go from any existing formula/algorithm to its "optimized" QC equivalent. I don't know if that can be done in a general way, like how we can switch from the time domain to the frequency domain with a Fourier transform, or take the derivative of an equation. Maybe it's more like an integral where the solution must be "found" by composing known integrals from a dictionary using heuristics. But without those transforms, I see us being stuck in a QC winter for the foreseeable future. My words are imperfect, but I know precisely what I'm getting at here. I can't run with QC the way I would with any other programming language, and it really bothers me. To the point of, I'll probably have to actually learn it from first principles and write it up from a programmer's perspective! Someday.
- JieJie 4y ago
  “The remaining probability was distributed over 01 and 10, which should not be a part of the Bell state. As we discussed before, this phenomenon is due to errors inherent to the quantum processor. As the backend technology improves we expect to get better results.” Debugging sounds like a real bear.

dr_dshiv 4y ago

Useful applications don’t exist. They surely will, at some point, but not yet. That’s why they aren’t provided as examples.

jtc42 4y ago
There are/seem to be very specific scientific applications, mostly around simulation of other systems whose hamiltonians map nicely to that of the computer. For almost everyone on the planet that isn’t very useful, but for that small handful it can be a game changer. … doesn’t contribute much here, but worth knowing! It’s exciting either way.
amelius 4y ago
Protein folding (computational chemistry), and breaking everyone's crypto.
- nabla9 4y ago
  > breaking everyone's crypto Symmetric encryption and hash functions with 256 bits are already quantum proof. Post-quantum algorithms are here in few years. Some of them are already in use in critical applications.
  - t_mann 4y ago
    Importantly, Bitcoin and Ethereum are (inadvertently) almost quantum-proof (because the addresses are hashes of the public keys), just not quite (because the public keys still get exposed during transactions). But at least the latter should be fixable (because the community is quite used to changes to the protocol anyway).
go_elmo 4y ago
I have to object, while not many algorithms exists, one of the most famous ones, Grovers algorithm [0] is also one of the most general ones and quite powerful. It reduces search of 2^n items to about 2^(n/2) while using n q-bits. Thats just the square root of a classical algorithm. [0] https://en.wikipedia.org/wiki/Grover%27s_algorithm https://en.wikipedia.org/wiki/Grover%27s_algorithm
latenightcoding 4y ago
of course they exist, we just don't have QCs with enough logical qubits yet.

nabla9 4y ago

> How do I use this to speed up database access? How do I use this to better commute graphics? Can you make a neural net better with this? It seems that you have missed some of the basics of quantum computing. For quantum computing to work the system must be isolated from the environment for the period of the computation. You can't use it for anything interactive like graphics. Any practical problem worth solving with quantum computer needs something like years or millions of years to solve using classical computer. Do you have database index where search takes years or graphics problem where output takes years? Also consider the amount of qbits needed. For database search you need quantum computer with same order of magnitude qbits as the database index. Practical problems for quantum computing are mostly scientific problems like simulating chemistry.

keithalewis 4y ago

Factoring large integers in order to decrypt RSA. Quantum computations do not produce numbers, only probability distributions. This is fine if all you are trying to do is factor integers since the answer can be easily verified. This is why massive amounts of money is being thrown at quantum computing.

packetlost 4y ago

QPUs are effectively (and theoretically) high-speed, specialized coprocessors at best. There's no way they're going to be doing anything you mentioned any time soon because that's not what they're good at or built for.

zmgsabst 4y ago
There’s some research on AI, eg at Microsoft Research. https://arxiv.org/abs/1412.3489 https://arxiv.org/abs/1412.3489

t_mann 4y ago

You're asking whether they can do things that classical computers can do reasonably well already marginally better, while their whole point is that they (one day, hopefully) can do things that current computers can't do at all (at scale, at least). Most of the speculative applications I've heard are in science, cryptography and AI/ML.

HWR_14 4y ago

IBM has (had?) a quantum course? Was it a free online course? And if one wants to take it for curiosity and not expecting to use the skills (immediately although maybe in the future) would you recommend it?

go_elmo 4y ago

I have to object, while not many algorithms exists, one of the most famous ones, Grovers algorithm [0] is also one of the most general ones and quite powerful. It reduces search of 2^n items to about 2^(n/2) while using n q-bits. Thats just the square root of a classical algorithm. [0] https://en.wikipedia.org/wiki/Grover%27s_algorithm https://en.wikipedia.org/wiki/Grover%27s_algorithm

zackmorris 4y ago

Ya looking at quantum computing (QC) from a programmer's perspective, I see issues with abstraction at multiple levels. Basically what's going on is that it's too narrow (tightly coupled to its implementation details), but also too broad (generalized over many disciplines) to easily transform it into simpler domains that are easier for humans to reason about. So for example, in computer science we can explain basic data structures like binary trees and hash tables from whichever context a student understands. We can use pointers and references from C-style imperative languages. We can use car and cdr from Lisp-style functional languages. We can even jump up to higher levels of abstraction with associative arrays in Javascript. Or make the whole thing in key-value stores like Redis or even a SQL database or x86 assembly if that floats our boat. But with QC, even its "assembly language" of Hilbert spaces and tensors is already so difficult to understand, much less apply, that the terminology feels like gatekeeping. There's no way to easily identify the smoking gun(s) which separate a QC simulation from qubits in the real world. It's like asking an assembly language teacher how a full adder works, but they throw up their hands and say that first we must understand Fermi levels and electron diffusion across a PN junction. Which while important, doesn't answer the student's question. What's needed are simple transforms to go from any existing formula/algorithm to its "optimized" QC equivalent. I don't know if that can be done in a general way, like how we can switch from the time domain to the frequency domain with a Fourier transform, or take the derivative of an equation. Maybe it's more like an integral where the solution must be "found" by composing known integrals from a dictionary using heuristics. But without those transforms, I see us being stuck in a QC winter for the foreseeable future. My words are imperfect, but I know precisely what I'm getting at here. I can't run with QC the way I would with any other programming language, and it really bothers me. To the point of, I'll probably have to actually learn it from first principles and write it up from a programmer's perspective! Someday.

JieJie 4y ago
“The remaining probability was distributed over 01 and 10, which should not be a part of the Bell state. As we discussed before, this phenomenon is due to errors inherent to the quantum processor. As the backend technology improves we expect to get better results.” Debugging sounds like a real bear.