Machine Learning on Near-Term Quantum Computers
TimeThursday, 21 November 201911am - 11:30am
DescriptionQuantum computers are known to solve certain problems that are otherwise intractable for the best classical algorithms. However, quantum devices in the near-term are still yet to develop capabilities that can yield practical advantages. A more fruitful direction for near-term quantum devices is then to use them as "sampling devices" which allow for efficient sampling from probability distributions that are arguably hard to simulate using classical computers.
One such hard sampling task exists in the training of Boltzmann Machine, which is a versatile model of probabilistic neural network that finds applications in a variety of areas in machine learning. In particular, accurately computing the gradient during the training of a restricted Boltzmann Machine requires extensive sampling from the model distribution, which is costly. We will present two different lines of work that have been pursued by our team, one using quantum annealers and the other using gate-model quantum devices.