Topic：An Information Centric-Approach to Data Analytics: The New Era to Information Theory, Machine Learning and Signal Processing

Date&Time：2016/09/23 (五) 14:20

Speaker：黃紹倫 博士

Location：清華大學台達館 R216

Abstract：
With the explosive growth of Internet and mobile communications, efficiently and effectively analyzing big amount data has emerged as an important issue.
In this talk, we present an information-centric approach to provide a theoretical foundation for mining critical knowledge of data. In particular, the information theoretic measurements, such as the mutual information and Kullback-Leibler (KL) divergence, are employed as the metric to select informative features from the empirical behaviors of the data.
In addition, we develop an efficient algorithm to compute informative feature functions and apply to two practical problems, namely, the Netflix and MNIST. We show that our approach has great performances in these two different scenarios, hence supplies a general methodology to deal with different kinds of engineering problems.