Keynote Speakers

Pavel Loskot
Zhejiang University | ZJU-UIUC INSTITUTE,
China.

Title: Model-Based vs. Sample-Based Data Processing Methods

Abstract: Mathematical models can often be developed to study and design various engineering systems. The models also enable devising optimum data processing strategies. There are, however, situations when the models are not known or only partially known. In such cases, an alternative strategy for developing data processing methods is to assume certain classes of universal models, which can be implicitly identified using a sufficient number of input-output training samples. The downside of considering universal models is their high computational, memory, and training costs, which are often also financial. In this talk, I will first argue that the models of systems can be equivalently described to varying degrees of accuracy using mathematical expressions, computer algorithms, or input-output pairs of samples. I will then explain how the supervised and unsupervised learning strategies have been used for decades to design adaptive data processing and filtering methods. Finally, I will conclude my talk by briefly outlining the recent developments in using universal filters, which are based on deep learning architectures.

Simon Fong
University of Macau
Macau.

---

This page will be further updated