2020 International Conference On Computer Aided Design

The Premier Conference Devoted to Technical Innovations in Electronic Design Automation

November 2-5, 2020San Diego Mission Bay Resort

MP Associates, Inc.

MONDAY November 04, 1:45pm - 3:45pm | Westminster II

Approximation in Behavioral Specifications, Neural Network Design, and Stochastic Computing
Hai Zhou - Northwestern Univ.
Approximation as a means for optimizing power, energy, and security in systems is addressed in the papers in this session. The first paper introduces the notion of approximation within the high-level synthesis flow. The second paper proposes a method for incremental training of neural networks using approximate multipliers. The third paper presents a technique for design space exploration for approximate neural networks without expensive retraining. Finally, the last paper of the session introduces a method for leveraging randomness in designing secure stochastic systems.

2C.1Approximating Behavioral HW Accelerators Through Selective Partial Extractions Onto Synthesizable Predictive Models
 Speaker: Siyuan Xu - Univ. of Texas at Dallas
 Authors: Siyuan Xu - Univ. of Texas at Dallas
Benjamin Carrion Schaefer - Univ. of Texas at Dallas
2C.2INA: Incremental Network Approximation Method for Limited Precision Deep Neural Networks
 Speaker: Zheyu Liu - Tsinghua Univ.
 Authors: Zheyu Liu - Tsinghua Univ.
Kaige Jia - Tsinghua Univ.
Weiqiang Liu - Nanjing Univ. of Aeronautics and Astronautics
Qi Wei - Tsinghua Univ.
Fei Qiao - Tsinghua Univ.
Huazhong Yang - Tsinghua Univ.
2C.3ALWANN: Automatic Layer-Wise Approximation of Deep Neural Network Accelerators without Retraining
 Speaker: Vojtech Mrazek - Brno Univ. of Technology
 Authors: Vojtech Mrazek - Brno Univ. of Technology
Zdenek Vasicek - Brno Univ. of Technology
Lukas Sekanina - Brno Univ. of Technology
Muhammad Abdullah Hanif - Vienna Univ. of Technology
Muhammad Shafique - Vienna Univ. of Technology
2C.4Exploiting Randomness in Stochastic Computing
 Speaker: John Hayes - Univ. of Michigan
 Authors: Paishun Ting - Univ. of Michigan
John Hayes - Univ. of Michigan