Binary Multilayer Perceptron using Quantum Annealers
Information
Contributors:
Abstract:
Quantum computing is completely different from the von Neumann architecture computers, and its performance growth is unrelated to the Moore's Law. Therefore, quantum computing is beginning to prosper in recent years. Quantum annealing (QA) is a metaheuristic based on quantum computing. Some quantum computers for QA have been commercialized such as D-Wave 2000Q and quantum-inspired annealers such as Fujitsu Digital Annealer (DA). QA is expert in solving combinatorial optimization problems. To expand the application of QA, we proposed a QA-based training algorithm for a binary multilayer perceptron (MLP). Our main idea is mapping the weights of a binary MLP to the Ising spins on the QA. It can train the binary MLP through a single run of quantum annealing. We confirmed that a MLP model trained with our method can classify the subset of the IRIS dataset.
- Junya Arai (Nippon Telegraph and Telephone)
- Junji Teramoto (Nippon Telegraph and Telephone)
- DEMA BA (Nippon Telegraph and Telephone)
Abstract:
Quantum computing is completely different from the von Neumann architecture computers, and its performance growth is unrelated to the Moore's Law. Therefore, quantum computing is beginning to prosper in recent years. Quantum annealing (QA) is a metaheuristic based on quantum computing. Some quantum computers for QA have been commercialized such as D-Wave 2000Q and quantum-inspired annealers such as Fujitsu Digital Annealer (DA). QA is expert in solving combinatorial optimization problems. To expand the application of QA, we proposed a QA-based training algorithm for a binary multilayer perceptron (MLP). Our main idea is mapping the weights of a binary MLP to the Ising spins on the QA. It can train the binary MLP through a single run of quantum annealing. We confirmed that a MLP model trained with our method can classify the subset of the IRIS dataset.
