Please join us for Alex's PhD proposal seminar.

Open to Staff and Students only.
This seminar will also be available to attend online via zoom (ID: 914 4624 1018) Contact Caitlin at the School Office if you have any questions (caitlin.warwood@vuw.ac.nz).

Abstract

Big data along with the advent of computational power have made machine learning a standard practice for data-driven decision-making. However, in practice, there are constraints of using real data, where synthetic data can be used as a substitute and complement. Recently, various deep learning methods have been proposed for data synthesis in the domain of computer vision and natural language. However, when it comes to tabular data, this research is still at an early stage. Tabular data synthesis is not trivial due to the nature of tabular data where mixed data types exist, and the underlying distribution is extremely complex. Therefore, how to improve the efficiency of tabular data representation, generation and evaluation are the key research objectives in this study. This work aims to alleviate the tabular data limitations, ranging from quantity, quality, and availability, by developing a novel deep learning based tabular data synthesis framework. We expect the new developments to provide a potential data solution for further improving the performance of the state-of-art classification models, while partially remove the barrier of accessing confidential microdata for academic research. Finally, the proposed framework will be applied to data in reliability, healthcare, and government administration to make a genuine industry impact.