December 14, 2009 at 14:52Google taught a quantum computer to recognize images
Google researchers said they were able to get a quantum computer to recognize cars in photos. This was reported on the official blog of the company.
The work was carried out in collaboration with the Canadian company D-wave, which provided the Chimera chip for work. At present, however, far from all physicists agree that this chip is indeed a quantum computer in the sense in which theorists understand this term.
As part of the work, the researchers used the so-called adiabatic algorithm. The essence of this algorithm is as follows. Consider a system whose states are the solution to some well-known problem. Then, rather slowly and adiabatically (that is, without exchanging heat with the external environment), we “deform” the system to another, to which the studied problem corresponds. It is argued that the “deformed” states of the new system will represent the desired solutions.
Using a similar algorithm, scientists tried to “teach” the system to recognize cars in the picture. For this, 20,000 photos were fed to the algorithm, half of which were cars. Each of the vehicles was manually placed in a special frame. After that, the system was given to work with raw photos. As a result, the chip managed, according to scientists, faster than any of the Google systems. Scientists emphasize that the practical application of the new technology is still far away.
More recently, scientists have managed to create the first programmable quantum (in the "right" sense of the word) computer. The machine of scientists works with two qubits, which can simultaneously be in two states. In the new installation, these objects are realized as beryllium ions in a miniature (about 200 nanometers) magnetic trap.
add-ons from ALEX_A :
As far as I understand, a quantum computer is used here to train a neural network. For optimization, the Quantum Adiabatic Algorithm is used. If this is so, then this is a very serious breakthrough in the field of designing neural networks, and therefore the development of various AI systems, since now it will be much easier to achieve training of a complex neural network and it is even possible to find a global minimum of error (I'm not sure about that), without wandering endlessly at local minima.
Here are links to the news and explanations for it:
Google demonstrates quantum computer image search
Machine Learning with Quantum Algorithms
Quantum Adiabatic Algorithms, Small Gaps, and Different Paths
The work was carried out in collaboration with the Canadian company D-wave, which provided the Chimera chip for work. At present, however, far from all physicists agree that this chip is indeed a quantum computer in the sense in which theorists understand this term.
As part of the work, the researchers used the so-called adiabatic algorithm. The essence of this algorithm is as follows. Consider a system whose states are the solution to some well-known problem. Then, rather slowly and adiabatically (that is, without exchanging heat with the external environment), we “deform” the system to another, to which the studied problem corresponds. It is argued that the “deformed” states of the new system will represent the desired solutions.
Using a similar algorithm, scientists tried to “teach” the system to recognize cars in the picture. For this, 20,000 photos were fed to the algorithm, half of which were cars. Each of the vehicles was manually placed in a special frame. After that, the system was given to work with raw photos. As a result, the chip managed, according to scientists, faster than any of the Google systems. Scientists emphasize that the practical application of the new technology is still far away.
More recently, scientists have managed to create the first programmable quantum (in the "right" sense of the word) computer. The machine of scientists works with two qubits, which can simultaneously be in two states. In the new installation, these objects are realized as beryllium ions in a miniature (about 200 nanometers) magnetic trap.
add-ons from ALEX_A :
As far as I understand, a quantum computer is used here to train a neural network. For optimization, the Quantum Adiabatic Algorithm is used. If this is so, then this is a very serious breakthrough in the field of designing neural networks, and therefore the development of various AI systems, since now it will be much easier to achieve training of a complex neural network and it is even possible to find a global minimum of error (I'm not sure about that), without wandering endlessly at local minima.
Here are links to the news and explanations for it:
Google demonstrates quantum computer image search
Machine Learning with Quantum Algorithms
Quantum Adiabatic Algorithms, Small Gaps, and Different Paths