Dr. Badru Mochizuki and Dr. Yimeng Sun, Assistant Professors at Kyoto College of Computer Science and Graduate Institute of Information (KCGI), presented their work on November 11, 2020 at the ITU AI/ML in 5G Challenge, which aims to solve real-world problems by applying machine learning to communication networks, and were selected as one of the top three teams in Japan and won the Grand Prize.
Dr. Mochizuki and Dr. Sun selected "Network State Estimation by Video Analysis in Real-Time Streaming Service" from the three themes presented by the conference, and estimated the network state (throughput and loss rate) by using machine learning in real time.
The team's machine learning model achieved a recognition accuracy of 96.3% for network state estimation from a real-time video dataset using deep learning.
The model constructed by Dr. Mochizuki and Mr. Sun is based on a neural network, and
- (1) Lightweight model: low computation time (can estimate the network state of 500 videos in 1 second)
- Portable: Platform-independent
These advantages led to the awarding of the Grand Prize.
Machine learning is now the lifeline of our society, and this presentation proposes one of the ways to apply machine learning to future networks, which could be used to reduce communication speed, improve communication latency, and improve the communication system itself in the future.