Slashdot Mirror
Google Demonstrates Quantum Computer Image Search
An anonymous reader sends along this quote from New Scientist:
"Google's web services may be considered cutting edge, but they run in warehouses filled with conventional computers. Now the search giant has revealed it is investigating the use of quantum computers to run its next generation of faster applications. Writing on Google's research blog this week, Hartmut Neven, head of its image recognition team, reveals that the Californian firm has for three years been quietly developing a quantum computer that can identify particular objects in a database of stills or video (PDF). Google has been doing this, Neven says, with D-Wave, a Canadian firm that has developed an on-chip array of quantum bits — or qubits — encoded in magnetically coupled superconducting loops."
I trust Google not to do anything unbelievably stupid (a bit silly perhaps, but nothing too absurd) but thinking that D-Wave can make a quantum computer is a very, very bad idea. Now it sounds like Google has been working on the algorithm side and I suspect that they're doing good work. The trouble is that D-Wave is doing the hardware. This is a company that has yet to demonstrate any success whatsoever.
They frequently release press updates saying that they have added more bits to the machine but they have never shown it to work for even a small number of bits. The physicists who developed the idea of an adiabatic quantum computer say that D-Wave seems to have misinterpreted their theory to make unrealistic claims and the whole thing is regarded as a bit of a joke in the physics community.
That said, developing the algorithms is a worthwhile thing to do so Google may not be relying on D-Wave to justify their research. I hope not. D-Wave may actually be on to something big that they haven't revealed to the scientific community, but probably not.
So if this is the future...where's my jet pack?
if you did an undergraduate physics degree, I'd be surprised if you didn't know what all of those words mean. They can all be wikid (not sure I like that word)...
Nothing to see here.
The problem with D-Wave’s approach is that it is not clear how well it can scale. Their adiabatic strategy involves starting in the ground state of one physical system, transforming it into another system very slowly ( “adiabatically” == very slowly), and then hoping that they stay in the ground state all the way to the end of the procedure; if they succeed in this, then they can read out the new state and they have the answer that they want.
The problem is that this only works as long as it is hard for the system to bump itself up into an excited state. However, as you attack larger and larger problems, the “energy gap” between the ground state and the first excited state shrinks exponentially with the size of the problem, greatly increasing the probability that you won’t end up with the right answer at the end of the computation.
In order to get around this problem, you need to do two things. First, you need to cool the system down so that its temperature is less than the energy gap. However, D-Wave’s cooling system does not accomplish this --- their temperature is too high. In fact, they freely admit that their temperature is larger than the energy gap, it’s just that they are gambling that in practice they can get away with it.
Second, you need to run the transformation very slowly --- at a speed that is roughly proportionate to the size of the energy gap. This might also turn out to case problems for D-Wave as they start scaling up their system to attack useful problems. Furthermore, although they have demonstrated a case where their computer shows a speedup over classical algorithms, this should be taken with a great of salt because as I understand it they basically applied their algorithm in a case where conditions favored it. (Mind you, that isn’t in itself a bad thing --- it is good to understand the conditions under which an existing quantum computer can ever beat an existing classical computer; given the infancy status of the field, I amazed that this can be done at all!)
So in short: no, D-Wave is not a scam, but they are taking a gamble that certain theoretical problems will not bite them in practice, and most QC researches tend to believe that they will lose this gamble even though we hope that they will win it.
Snarkiness is inversely proportional to wisdom because it emphasizes feeling right rather than being right.