Professor Kyunghan Lee (Corresponding Author, Institute of New Media and Communications)'s team of Seoul National University College of Engineering's Department of Electrical and Computer Engineering, along with KAIST and University of Colorado Boulder won the Best Paper Award at ACM MobiSys 2021 (International Conference on Mobile Systems, Applications, and Services). As a Korean University, the research team was the first to win the Best Paper Award in the history of MobiSys, which began in 2003.
 
The award-winning paper, "zTT: Learning-based DVFS with Zero Thermal Throttling for Mobile Devices" involves solving the problem of rapid performance degradation due to thermal throttling, caused by excessive heating of mobile terminals such as 5G smartphones, through reinforcement learning-based Dynamic Voltage and Frequency Scaling (DVFS).
 
In order to increase the user's sensory performance while preventing rapid performance degradation due to thermal throttling, it is key to perform optimal power distribution between the CPU and GPU within the total power consumption range to maintain an appropriate temperature. However, since the allowable total power consumption range and optimal power distribution change in real time depending on the surrounding environment (ambient temperature, cooling situation, etc.) and application, it was recognized as a problem that is difficult to solve with the traditional DVFS technique. Professor Kyunghan Lee's team solved the problem of determining the total power and the optimal power distribution according to the surrounding environment and application by introducing a reinforcement learning technique that includes real-time learning. By strictly implementing the proposed techniques on mobile platforms, including smartphones, the team demonstrated that application performance can be significantly improved without generating thermal throttling (zero thermal throttling) and the results of the team's implementation are also certified through the ACM code review system. (ACM Results Reproduced Badge)
 
These results of Professor Lee's research team show that reinforcement learning-based system control can greatly contribute to overall system performance improvement even on mobile platforms, which was predicted to have the introduction of AI/ML techniques to be difficult due to power consumption problems, providing an opportunity to actively introduce AI/ML-based control techniques into next-generation operating systems.