Sequence Memorization in Dynamic & Quantum Boltzmann Machines
of Christoffer Arnlund
Date & Time: Nov 8, 13:00-14:00
Location: LUCAS (E:4130)
Memorizing sequences are useful for many applications such as Natural Lan- guage Processing and Anomaly detection but can also be used generative, e.g. to generate missing part of a photograph or music sheet.
We study the performance of memorizing sequences in Dynamic Boltzmann Machines (DyBM) and Quantum Boltzmann Machines (QBM). and develop a DyBM and QBM in Python 2.7. The hardware used for the QBM was the D- Wave 2000Q, a quantum annealing machine.
We train a Dynamic Boltzmann Machines with bitmap patterns of alpha- betical images and also train a Quantum Boltzmann machine with the goal of learning classification of the same alphabet used for the Dynamic Boltzmann Machine.
The results show that Dynamic Boltzmann Machines perform well for small data sets. The Quantum Boltzmann machine successfully classified the alphabet with ∼ 80 % accuracy.