Abstract:

Efficient deep learning is a broad concept that we aim to learn compressed deep models and develop training algorithms to improve the efficiency of model representations, data and label utilization, etc. In recent years, deep neural networks have been recognized as one of the most effective techniques for many learning tasks, also, in the foreseeable future the world will be populated with intelligent devices that require executable deep models on these inexpensive, low-power hardware platforms. Therefore, network compression including pruning, binarization, quantization, and their learning methods, together with the real-world applications will be an indispensable and new hot area in both academia and industry. It is also an interdisciplinary field that deals with the ability of how machines can be developed to obtain high-level understanding from data, meanwhile, significantly reducing the number of parameters and computational requirements of deep models.

In this talk, I will introduce my previous studies that are related to the efficient machine learning and its broad applications, covering recognition, detection, generative adversarial learning. My talk will focus on the following several aspects: 1) efficiency on storage and computational cost, including efficient training and inference by compressing networks, pruning, binarization, etc., as well as developing training strategies, such as knowledge distillation, learning detectors from scratch. 2) Data/label efficiency, including few-shot learning, self-supervised learning and their association with the compressed deep models.

BIO:

Zhiqiang Shen is currently a post-doctoral researcher in the ECE Department of CMU, working with Prof. Marios Savvides. His research interests span a broad area of machine learning, computer vision, etc. Prior to CMU, he was fortunate to be a joint-training Ph.D. student (2017-2019) in UIUC/IFP group, advised by Prof. Thomas S. Huang. He received his Ph.D. from Fudan University. He was a research intern at Intel Labs China, collaborating with Jianguo Li and Yurong Chen. He also serves as a reviewer for CVPR, ICCV, ECCV, NeurIPS, ICML, TPAMI, IJCV, etc.

Homepage:  http://zhiqiangshen.com/

Sponsored in part by:   Facebook Reality Labs Pittsburgh