THURSDAY July 23, 3:30pm - 5:00pm
TOPIC AREA: MACHINE LEARNING/AI, DESIGN
KEYWORD: ARCHITECTURE & SYSTEM DESIGN
EVENT TYPE: RESEARCH REVIEWED
Pruning and Compression for Emerging Deep Neural Networks
Chair:
Bipin Rajendran - King's College London, United Kingdom
Co-Chair:
Bert Moons - Qualcomm Research, Amsterdam, Netherlands Antilles
While algorithmic advances have substantially pushed the capabilities of deep learning, being able to run such complex algorithms on embedded hardware or edge devices has been challenging. For emerging deep neural networks featuring in-memory computing, capsule networks, intermittent inference, conditional computing, etc., effective pruning and compression techniques are presented in this session.

95.1PIM-Prune: Fine-Grain DCNN Pruning for Crossbar-Based Process-In-Memory Architecture
 Speaker: Li Jiang - Shanghai Jiao Tong Univ., Shanghai, China
 Authors: Chaoqun Chu - Shanghai Jiao Tong Univ., Shanghai, China
Yanzhi Wang - Northeastern Univ. , Boston, MA
Yilong Zhao - Shanghai Jiao Tong Univ., Shanghai, China
Xiaolong Ma - Northeastern Univ. , Boston, MA
Shaokai Ye - Tsinghua Univ., Beijing, China
Yunyan Hong - Shanghai Jiao Tong Univ., Shanghai, China
Yinhe Han - Chinese Academy of Sciences & Univ. of Chinese Academy of Sciences, Beijing, China
Li Jiang - Shanghai Jiao Tong Univ., Shanghai, China
95.2Tight Compression: Compressing CNN Model Tightly Through Unstructured Pruning and Simulated Annealing Based Permutation
 Speaker: Xizi Chen - Hong Kong Univ. of Science and Technology, Kowloon, Hong Kong, Hong Kong
 Authors: Xizi Chen - Hong Kong Univ. of Science and Technology, Kowloon, Hong Kong, Hong Kong
Jingyang Zhu - NVIDIA Corp., Shanghai, China
Jingbo Jiang - Hong Kong Univ. of Science and Technology, Kowloon, Hong Kong, Hong Kong
Chi-Ying Tsui - Hong Kong Univ. of Science and Technology, Kowloon, Hong Kong, Hong Kong
95.3Q-CapsNets: A Specialized Framework for Quantizing Capsule Networks
 Speaker: Beatrice Bussolino - Politecnico di Torino, Torino, Italy
 Authors: Alberto Marchisio - Vienna Univ. of Technology, Vienna, Austria
Beatrice Bussolino - Politecnico di Torino, Torino, Italy
Alessio Colucci - Technische Univ. Wien
Maurizio Martina - Politecnico di Torino, Torino, Italy
Guido . Masera - Politecnico di Torino, Torino, Italy
Muhammad Shafique - Technische Univ. Wien, Wien, Austria
95.4Intermittent Inference with Nonuniformly Compressed Multi-Exit Neural Network for Energy Harvesting Powered Devices
 Speaker: Yawen Wu - Univ. of Pittsburgh, Pittsburgh, PA
 Authors: Yawen Wu - Univ. of Pittsburgh, Pittsburgh, PA
Zhepeng Wang - Univ. of Pittsburgh, Pittsburgh, PA
Zhenge Jia - Univ. of Pittsburgh, Pittsburgh, PA
Yiyu Shi - Univ. of Notre Dame, Notre Dame, IN
Jingtong Hu - Univ. of Pittsburgh, Pittsburgh, PA
95.5High PE Utilization CNN Accelerator with Channel Fusion Supporting Pattern-Compressed Sparse Neural Networks
 Speaker: Jingyu Wang - Tsinghua Univ., Beijing, China
 Authors: Jingyu Wang - Tsinghua Univ., Beijing, China
Songming Yu - Tsinghua Univ., Beijing, China
Jinshan Yue - Tsinghua Univ., Beijing, China
Zhe Yuan - Tsinghua Univ., Beijing, China
Zhuqing Yuan - Tsinghua Univ., Beijing, China
Huazhong Yang - Tsinghua Univ., Beijing, China
Xueqing Li - Tsinghua Univ., Beijing, China
Yongpan Liu - Tsinghua Univ., Beijing, China
95.6BPNet: Branch-pruned Conditional Neural Network for Systematic Time-accuracy Tradeoff
 Speaker: Youngmin Yi - Univ. of Seoul, Republic of Korea
 Authors: Kyungchul Park - Samsung Electronics Co., Ltd., Seoul, Republic of Korea
Chanyoung Oh - Univ. of Seoul, Republic of Korea
Youngmin Yi - Univ. of Seoul, Republic of Korea