Online Data Imputation by Modified Mixture Density Networks

In the era of big data, software based on artificial intelligence has greatly improved the quality of people’s lives. Although data is not a scarce resource, the data we collected are usually incomplete due to many reasons, i.e. they contain some missing values. Simply deleting these missing data will not only cause great waste, but may also make trained models biased. To address this problem, data imputation algorithms can predict the missing values based on all the observed data, making the data set complete, so that the collected data can be used more efficiently. In the meanwhile, based on the same idea, we can apply this algorithm to data compression. We can estimate how much information is lost during the compression process, and evaluate whether the compressed data retains enough information for further usage.

Faculty Supervisor:

Linglong Kong

Student:

Partner:

NTwist

Discipline:

Mathematics

Sector:

Information and cultural industries; Manufacturing; Mining; Professional, scientific and technical services

University:

University of Alberta

Program: