Images, video, speech, audio, point clouds, light and sound fields… useful information is buried in these signals. We study and develop methods to efficiently sample and represent these signals, remove noise from them, compress them, and recover their missing pieces.
With the likes of Netflix, Spotify, and Skype, multimedia communications have become mainstream. But achieving dependable real-time transfer of multimedia signals is still a challenge, especially in the case of demanding immersive applications. We create techniques for error-resilient coding, power and resource management, and error control, to enable efficient and reliable multimedia communications.
The word ergonomics probably makes you think about comfortable chairs or pillows. But physical objects are not the only things we interact with. You likely spend a good portion of each day looking at images or video, listening to music, or browsing websites. Are these digital objects comfortable? We want to understand how people interact with digital objects, especially multimedia signals, in order to facilitate better user experience and more seamless interaction with our digital environment.
Deep Learning is revolutionizing the multimedia industry. We employ various forms of machine/deep learning to derive useful information from multimedia signals, even in their compressed form. We also study ways to deploy deep models efficiently, through collaborative intelligence.