Implementation of a convolutional layer using approximate multipliers in FPGA for convolutional neural networks

• Grant number: 18/00096-1
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Effective date (Start): April 01, 2018
• Effective date (End): July 31, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Ricardo Menotti
• Grantee: Leonardo Tavares Oliveira
• Supervisor: Nader Bagherzadeh
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil
• Research place: University of California, Irvine (UC Irvine), United States
• Associated to the scholarship: 17/13520-3 - An accelerator for deep neural convolutional networks implemented in FPGA, BP.IC

Abstract

The evolution of High Performance Computing (HPC) allowed the popularization of many programs and algorithms that required high computational power, such as Convolutional Neural Networks (CNN). However, the requirement for higher accuracy in classification increased the number of layers used by these networks, thereafter raising the power consumption of the whole system. Min Soo Kim's Ph.D. project, advised by Prof. Nader Bagherzadeh at University of California, Irvine, aims to reduce the aforesaid power consumption by using approximate computation for inference stage, while maintaining the accuracy levels. An approximate multiplier was proposed by Min Soo Kim, reducing the power consumption by 76.6% and achieving the same accuracy when compared to a fixed-point multiplier. This projects aims to implement a primitive convolutional layer in FPGA using the approximate multiplier proposed by Min Soo Kim, surveying and comparing the results with other approximate multipliers. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

KIM, MIN SOO; DEL BARRIO, ALBERTO A.; OLIVEIRA, LEONARDO TAVARES; HERMIDA, ROMAN; BAGHERZADEH, NADER. Efficient Mitchell's Approximate Log Multipliers for Convolutional Neural Networks. IEEE TRANSACTIONS ON COMPUTERS, v. 68, n. 5, p. 660-675, MAY 2019. (18/00096-1)