To enable the transcript, please select a language in the video player settings menu.
About this video
Generative models are usually memory and computation intensive. Compression techniques, such as quantization and pruning, enable a reduction of the model size and complexity, which may enable deployment on edge devices. Depending on the desired compression levels, the quality and diversity of the generated set may be affected in different ways. Hence, an accurate assessment of the compressed generative model is important for a given target application and data domain. More information...
Gonçalo Mordido is a Ph.D. student at Hasso Plattner Institute, working on the improvement of deep learning models. More specifically, his research focuses on generative adversarial networks, model compression, and evaluation of generative models. During his doctoral studies, he has interned at NVIDIA twice, having received the NVIDIA Recognition Award for his research contributions. Gonçalo obtained his bachelor's and master's degrees in Computer Science Engineering at the New University of Lisbon in 2015 and 2017, respectively.