Deshani Geethika Poddenige

Thesis Title

Towards Robust Use of Generative Adversarial Networks (GANs) with Neural Architecture Search (NAS)

Research overview

NAS focuses on exploring the joint optimization of possible wiring patterns and operations in neural networks. Since NAS uses a constrained search space of wiring patterns, novel neural architectures cannot be explored beyond the provided search space. In this study, we will focus on developing a GAN to generate novel neural architectures. Moreover, since GAN networks are human-designed, only a handful of GAN architecture types are found in the current literature. In this study, we focus on two main objectives; (1) using GANs to generate novel neural architectures and (2) alleviating human bias in designing GANs by automating their design.

Supervisors

Prof Saman Halgamuge

Dr Damith Senanayake

Qualifications

B.Sc. Computer Science and Engineering, University of Moratuwa, Sri Lanka (2018)