Fundamental limits of data transmission and storage

Noise always exists in our information processing systems such as communication, storage, and learning systems. Therefore, it is necessary to know the fundamental performance limits of such systems and how to attain them, which we investigate from an information-theoretic perspective. Typical topics and target applications include wireless communications (5G, 6G, etc) and data storage (JPEG, MPEG, etc).

Efficient/scalable ML

The rapid development of data collection methods enables us to obtain large volumes of high-dimensional data. Moreover, the success of overparameterized neural networks requires ML algorithms to be efficient and scalable. There are several approaches addressing this challenge, such as implementing ML algorithms in a distributed manner and pre-/post-processing data and inferences. We explore this topic from a theoretical perspective.

Interaction among decision makers

Our decision-making process (e.g., inference, classification, prediction, etc.) is based on our own data and observations of others' decisions. It motivates the study of social learning---essentially the study of the interaction of decision makers in a community. The target applications are not limited to social networks; they include any decision-making networks such as networks of classifiers, predictors, sensors, and/or human-machine mixed nodes. The overall goal is to understand and improve the learning process over networks.