Innovative algorithms for 6G vision services improve performance and save energy at the same time

Professor Jeongho Kwak from the DGIST Department of Electrical Engineering and Computer Science has developed a learning model and resource optimization technology that combines accuracy and efficiency for 6G vision services. This technology is expected to be utilized to address the high level of computing power and complex learning models required for 6G vision services.

6G mobile vision services are associated with innovative technologies such as augmented reality (AR) and autonomous driving, which are attracting great attention in modern society. These services allow you to quickly capture videos and images and efficiently understand their content through deep learning-based models.

However, this requires a high-performance processor (GPU) and an accurate learning model. Previous technologies treated learning models and computing/networking resources as separate entities, failing to optimize performance and resource utilization on mobile devices.

To address this problem, Professor Jeongho Kwak’s team focused on optimizing learning models and computing/network resources simultaneously in real-time. As a result, they developed a new integrated learning model and computing/computing model that can reduce energy consumption by at least 30% while maintaining average accuracy compared to current techniques, without sacrificing average target accuracy or time delay. We proposed a networking optimization algorithm, VisionScaling.

The VisionScaling algorithm developed by Prof. Kwaks’ team uses one of the latest learning techniques, Online Convex Optimization (OCO), to adapt to constantly changing mobile environments without prior knowledge of future conditions. Adapt to maintain optimal performance.

Additionally, Professor Kwak’s team implemented and tested a real-world mobile vision service environment using embedded AI devices and connected edge computing platforms. They confirmed that the developed VisionScaling algorithm saves 30% more energy on mobile devices and improves end-to-end latency by 39% compared to the previously used algorithm.

Professor Jeongho Kwak from the Department of Electrical Engineering and Computer Science at DGIST said, “This research makes practical contributions in implementing and verifying performance in irregularly changing mobile environments, and utilizes dynamic optimization and learning techniques. It fulfills both the mathematical contribution of proving optimal performance.” This is important because it provides the technical foundation for deep learning-based mobile services that require more memory/computing resources in the future. ”

This research IEEE Internet of Things Journal.