Adventures in Approximate Computing

Monday, 2 November 2015 at 14:00 in room INF 328

Marilyn Wolf, Georgia Institute of Technology, Atlanta, Georgia, USA

 

Abstract:

Approximate computing has been advanced as a methodology to more efficiently use computational resources in a post-Moore’s Law world.  This talk describes our work, joint with Se Hun Kim and Saibal Mukhopodhyay, on approximate computing for image compression and the lessons we learned. Perceptual models of visual information suggest the types of errors that can be tolerated in visual information; based on those models, we decided to explore subthreshold logic for the discrete cosine transform (DCT), a critical step in JPEG compression. Subthreshold logic is of interest for its very low energy operation. We developed an improved theory for error analysis of subthreshold logic operation. We used this to develop a subthreshold logic module for the DCT. However, our results showed that net energy gains in the DCT were offset by energy losses in the memory system due to poor compression. We developed alternative approximate computing algorithms based on quantization that provide significant system-level energy savings.
 

About the speaker:

Marilyn Wolf is Rhesa "Ray" S. Farmer Distinguished Chair of Embedded Computing Systems and Georgia Research Alliance Eminient Scholar at the Georgia Institute of Technology.  She received her BS, MS, and PhD in electrical engineering from Stanford University in 1980, 1981, and 1984, respectively.  She was with AT&T Bell Laboratories from 1984 to 1989.  She was on the faculty of Princeton University from 1989 to 2007.  Her research interests included embedded computing, embedded video and computer vision, and VLSI systems. She has received the ASEE Terman Award and IEEE Circuits and Systems Society Education Award. She is a Fellow of the IEEE and ACM and an IEEE Computer Society Golden Core member.