Implementing quick network
recovery from disasters
Telecommunication networks are now an important infrastructure that connects communities, individuals, and society. Therefore, when a disaster occurs in Japan, damaged base stations and cables must be restored as quickly as possible to restore access to social functions in communities. What is the best way to deploy personnel and distribute resources such as fuel and supplies to quickly restore the telecommunication network?
In an environment that allows him to focus on the basic areas rather than on making hasty practical applications, Wang Zhao is working on disaster simulation and optimization of recovery plans using deep reinforcement learning.
Deep reinforcement earning for disaster recovery planning
“As a researcher, it is important to be able to think freely. Without that, you will not be able to produce good research results. I believe it is very important to be in an environment where you can demonstrate your originality.” Wang Zhao, who studied in China until university and then entered a graduate school in Japan to pursue research, chose NTT R&D for being a place where he could demonstrate his originality in conducting his research. He is now working on the effective use of deep reinforcement learning for disaster recovery. Wang's research is to determine how to distribute resources to enable quick recovery using neural networks to prepare for possible disasters and other future events.
In disaster recovery, there are many factors to be considered, including not only the damage situation, but also where and how much energy will be used, how many people are affected, how much communication needs will arise, and how congestion will quickly occur and communication quality will deteriorate if users flood in. Although mathematical methods have been traditionally used to solve such combinatorial problems as resource allocation, it is difficult to take into consideration an immense number of parameters in real time using conventional methods.
“There are many possible disaster situations. It is difficult to deal with them using conventional optimization methods. Therefore, I believe that calculating with deep reinforcement learning using neural networks will make it possible to show calculation results that will allow quick recovery in the field.”
Our mission is to maintain telecommunication services.
NTT possesses a large amount of data on past telecommunication events. Wang's research has been conducted in cooperation with NTT West, collecting information on the company’s communication buildings and ranking the buildings in accordance with disaster simulations, among other field-based research activities. Wang feels strongly optimistic about being able to conduct cutting-edge deep reinforcement learning research in an environment where he can handle data rooted in real-world operations.
“I am conducting my research in a way that allows me to present papers at conferences and international meetings several times a year. However, it is important to go beyond the academic aspects and link my research to practical applications. If I don’t envision the practical applications, it will end up being a purely academic pursuit. I think both the academic aspect and the engineering aspect focused on practical applications are important.”
The algorithms Wang is researching are likely to be of great benefit not only in times of disaster, but also for use in maintenance work during normal times.
“Communication is a social infrastructure. It is very important to be able to communicate with businesses, hospitals, and family members not only during normal times, but also in times of disaster. That is why I believe that our mission is to restore communication services as quickly as possible, and I am happy if I can contribute to that end.”
Pressure can also be a driving force for effort.
Around the time I entered the doctoral program, I started to think about moving on to an environment where I could conduct stable, long-term basic research. However, the area of deep learning that I had been involved in often required producing immediate business applications, and I felt that there were few places where I could sit back and engage in basic research. I then learned that NTT R&D conducts active research in a variety of fields and layers, and comprehensive research is encouraged. My graduate school supervisor was from NTT R&D, and he recommended NTT R&D to me, saying that it is a good place to work, which also led me to choose NTT R&D.
After joining the company, when I finally started my research activities, I remember being surprised at the high caliber of the people around me. Many are presenting their work at top conferences, and the level of discussions in everyday casual conversations is also very high and very stimulating. Everyone produces excellent results every year, and to be honest, I sometimes feel pressure; but that pressure also serves as a driving force for me to work harder.
Also, at international conferences, I often have the opportunity to meet up with friends I studied with as an undergraduate in China. I think that being in an environment where I can attend leading-edge conferences and proudly present my achievements is a source of much energy for me as a researcher.
- Network Service Systems Laboratories. Wang Zhao Joined the company in 2018
- Wang, who has Chinese parents, lived in the U.S., speaks three languages, and is currently working in Japan, possesses a truly "cosmopolitan" sense of values. He feels that the free atmosphere at NTT R&D suits him well. “If you are looking for a free research environment, want to be a leading-edge researcher, want to work with leading-edge researchers, or want to keep working on your research theme over the long term, come to NTT R&D!”
※The names of research institutes and affiliate organizations of employees in the article are current at the time of the interview. Some may be old names of research institutes.
