Presentation
Multi-class Support Vector Machine on Quantum Annealers
SessionResearch Poster Session
Event Type
Research Poster
Pre-Recorded
TimeTuesday, June 23rd4:15pm - 4:20pm
LocationAnalog 1
DescriptionThe end of Moore's Law makes it difficult for traditional silicon-based computers to attain significant improvements in computing performance. Therefore, new generation computing such as quantum computing is beginning to prosper. Both universal gate quantum computing and quantum annealing are considered to have the possibility of yielding significant improvements in computing performance. Quantum annealing was specifically designed for combinatorial optimization problem and is easier to realize than universal gate quantum computing. These benefits have encouraged the commercialization of quantum annealers such as D-Wave 2000 and quantum-inspired annealers such as Fujitsu Digital Annealer (DA).;
To expand the application area of quantum annealing, some studies have focused on how to use quantum annealing for other problems such as binary quantum SVM, which is one of the most famous algorithms in machine learning. However, in practical applications, there are more multi-class problems than binary ones.;
In this study we propose a multi-class SVM algorithm that can be used by a quantum annealer. Our main idea is to compare the energy values obtained by quantum annealing of multiple binary classifiers to find the largest margin for each class. We evaluate our method on quantum annealer and simulated annealing using a 3-class synthetic dataset and a benchmark dataset (IRIS). The results show that our method can classify multi-class data with precision comparable to that offered by existing alternatives.
To expand the application area of quantum annealing, some studies have focused on how to use quantum annealing for other problems such as binary quantum SVM, which is one of the most famous algorithms in machine learning. However, in practical applications, there are more multi-class problems than binary ones.;
In this study we propose a multi-class SVM algorithm that can be used by a quantum annealer. Our main idea is to compare the energy values obtained by quantum annealing of multiple binary classifiers to find the largest margin for each class. We evaluate our method on quantum annealer and simulated annealing using a 3-class synthetic dataset and a benchmark dataset (IRIS). The results show that our method can classify multi-class data with precision comparable to that offered by existing alternatives.
Poster PDF