2019 International Conference On Computer Aided Design

The Premier Conference Devoted to Technical Innovations in Electronic Design Automation

November 4-7, 2019The Westin Westminster Westminster, CO

MP Associates, Inc.
TUESDAY November 05, 10:30am - 12:00pm | Westminster III
EVENT TYPE: EMBEDDED TUTORIAL
SESSION 5D
When Neural Networks Meet Hardware: The Princess, The Knight, and the Very Bad Dragon
Moderator:
Qinru Qiu - Syracuse Univ.
Organizer:
Jingtong Hu - Univ. of Pittsburgh
Hardware acceleration of neural networks, which combines the performance of neural networks and the speed of dedicated hardware, has been widely adopted in many applications. However, the marriage between hardware and neural networks is not as straightforward as it seems. Careless implementations without considering various hardware related issues would easily lead to designs with inferior speed, performance, or even reliability. In this tutorial, three talks will cover the most critical parts in the design flow of hardware accelerators of neural networks. Specifically, the first talk will discuss hardware-aware neural architecture search; The second talk will introduce hardware-aware neural architecture compression; The third talk will cover hardware-defect-tolerant neural networks. Together they convey an important message that neural network and hardware needs to be co-designed and co-optimized throughout the design flow for maximum speed, performance and reliabilty.

5D.1Hardware-aware Neural Architecture Search
 Speaker: Qing Lu - Univ. of Notre Dame
 Authors: Qing Lu - Univ. of Notre Dame
Weiwen Jiang - Univ. of Notre Dame
Jingtong Hu - Univ. of Pittsburgh
Yiyu Shi - Univ. of Notre Dame
5D.2How to Obtain and Run Light and Efficient Deep Learning Networks
 Speaker: Fan Chen - Duke Univ.
 Authors: Fan Chen - Duke Univ.
Wei Wen - Duke Univ.
Linghao Song - Duke Univ.
Jingchi Zhang - Duke Univ.
Hai Li - Duke Univ.
Yiran Chen - Duke Univ.
5D.3Making the Fault-Tolerance of Emerging Neural Network Accelerators Scalable
 Speaker: Tao Liu - Florida International Univ.
 Authors: Tao Liu - Florida International Univ.
Wujie Wen - Lehigh University