EE Seminar: DEGAS: Differentiable Efficient Generator Search
Speaker: Sivan Doveh
M.Sc. student under the supervision of Dr. Raja Giryes
Wednesday, March 11th, 2020 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering
DEGAS: Differentiable Efficient Generator Search
Abstract
Network architecture search (NAS) achieves state-of-the-art results in various tasks such as classification and semantic segmentation. Recently, a reinforcement learning-based approach has been proposed for Generative Adversarial Networks (GANs) search.
In this work, we propose an alternative strategy for GAN search by using a method called DEGAS (Differentiable Efficient GenerAtor Search), which focuses on efficiently finding the generator in the GAN.
Our search algorithm is inspired by the differential architecture search strategy and the Global Latent Optimization (GLO) procedure. This leads to both an efficient and stable GAN search.
After the generator architecture is found, it can be plugged into any existing framework for GAN training. For CTGAN, which we use in this work, the new model outperforms the original inception score. It also gets better results than the RL based GAN search methods in shorter search time.
