CPRE 4870: Hardware Design for Machine Learning
(Dual-listed with CPRE 5870).
Credits: 4. Contact Hours: Lecture 3, Laboratory 3.
Prereq: COMS 3210 or CPRE 3810
Introduction to hardware architectures for machine learning. Full system view – machinelearning frameworks to hardware interface to hardware architecture. General purpose CPU extensions for machine learning. GPU extensions for machine learning. Spatial architec-tures for machine learning. Performance, energy, and accuracy trade-offs. Hardware designoptimizations for machine learning, including quantization, data re-use, SIMD, and SIMT. Lab section will culminate with the design and evaluation of an application-specific machinelearning accelerator.