### Adaptive Hyperparameter Updating for Training Restricted Boltzmann Machines on Quantum Annealers (Jan. 28, 2021)

• Published: 2021-01-25

Time: 9:00-10:30am, Jan 28 (Thur.), 2021  (Beijing/Shanghai)

Unsupervised machine learning via training neural networks has a variety of applications such as image recognition, drug discovery, and materials design. One of commonly used the techniques is called Restricted Boltzmann Machines (RBMs). The RBMs' Boltzmann probability distribution is used as a model to identify network parameters. Training such networks needs to calculate ensemble expectation values of corresponding model, and often requires large computational power even search a large probability space for estimation results. Quantum annealing has been proposed as a means to search this space more efficiently which has been experimentally investigated on D-Wave hardware. D-Wave implementation requires selection of a hyperparameter ($\beta$) within the Boltzmann distribution which can strongly influence optimization. Previous researches often take this hyperparameter a predefined constant or estimate an effective temperature which requires additional sampling process. Here we proposed two methods that adaptively updating this hyperparameter utilizing hardware implementation data. We find both methods improve training RBMs based upon D-Wave hardware experimental validation on an image recognition problem.