Predavanje: "Approximate GEMM unit for High-Performance computing"

Sažetak predavanja:

Approximate computing has emerged as a popular strategy for energy-efficient circuit design, where the challenge is to achieve the best tradeoff between design efficiency and accuracy. The essential operation in artificial intelligence algorithms is the general matrix multiplication (GEMM) operation comprised of matrix multiplication and accumulation. An approximate general matrix multiplication (AGEMM) unit employs approximate multipliers to perform matrix-matrix operations on four-by-four matrices given in sixteen-bit signed fixed-point format. The synthesis of the proposed AGEMM unit to the 45 nm Nangate Open Cell Library revealed that it consumed only up to 36% of the area and 25% of the energy required by the exact general matrix multiplication unit. The AGEMM unit is ideally suited to convolutional neural networks, which can adapt to the error induced in the computation. The results on honeybee detection with the YOLOv4-tiny convolutional neural network imply that we can deploy the AGEMM units in convolutional neural networks without noticeable performance degradation. Moreover, the AGEMM unit’s employment can lead to more area- and energy-efficient convolutional neural network processing, let it be in the edge devices or HPC centres.

O predavaču:

Dr Ratko Pilipović is currently employed as a research and teaching assistant at the Faculty of Computer and Information Science, University of Ljubljana, Slovenia and is a member of the Laboratory for Parallel processing and Adaptive Systems. Before he was a teaching assistant at the Faculty of Electrical Engineering, University of Banjaluka. In October 2021, he obtained his PhD from the Faculty of Computer and Information Science, University of Ljubljana, Slovenia.  His research interests include approximate computing, arithmetic circuit design, FPGA design, embedded processing and machine vision.