Researchers develop deep learning-based video streaming technology

Park Sae-jin Reporter Posted : 2018-10-30 14:30 Updated : 2018-10-30 14:30
글씨작게 글씨크게

[The Korea Advanced Institute of Science & Technology]

SEOUL -- South Korean researchers have developed a new deep learning-based technology that will enable viewers to watch high-quality video content in an unfavorable situation such as slow internet speed.

The Korea Advanced Institute of Science & Technology (KAIST), a prominent state research institute, said that its technology uses a deep neural network (DNN) to enhance the quality of video regardless of available bandwidth. 

DNN is optimized for deep learning, a class of machine learning algorithms for artificial intelligence (AI). Online video services like YouTube use adaptive streaming to automatically send low-quality video data when the network's speed is slow.

Video streaming services transfer small data segments one by one instead of sending out a whole video file to prevent traffic overload and automatically select video quality. 

KAIST researchers combined adaptive streaming and a deep learning-based image enhancing technique to create the so-called Neural Adaptive Content-aware Internet Video Delivery technology. When small packets of data are received, a DNN processor improves the quality of videos. 
The video quality enhancing technology was applied to desktops and KAIST plans to apply it to portable devices. The results of the research were presented at the 13th USENIX Symposium held in the United States on October 10.