Events Calendar

Electrical Engineering Seminar

Speaker: Swagath Venkataramani, Purdue University

Talk Title: Addressing the Efficiency Gap with Approximate Computing

Abstract: The “efficiency gap” created by diminishing benefits from semiconductor technology scaling on the one hand, and projected growth in computing and data demand on the other, has created an urgent need to identify new sources of computing efficiency across the computing stack. Fortunately, the workloads that drive the demand for computing efficiency also present new opportunities. In data centers and the cloud, the demand for computing is driven by the need to organize, search through, analyze, and draw inferences from, exploding amounts of digital data. In mobile and embedded devices, the need to more naturally and intelligently interact with the physical world, and process richer media drive much of the computing demand. A common pattern that emerges from both ends of the spectrum is that these applications are largely not about calculating a precise numerical answer; instead, “correctness” is defined as producing results that are good enough, or of sufficient quality, to produce an acceptable user experience. As a result, these workloads are endowed with a high degree of intrinsic resilience to their underlying computations being executed in an approximate or inexact manner. Approximate computing broadly refers to exploiting the forgiving nature (or intrinsic resilience) of applications to design more efficient (faster, lower power) computing platforms. In this talk, I will describe how current workload trends are driving interest in approximate computing, and describe a vision for approximate computing at all layers of the computing stack. To realize this vision, I will outline a holistic approach that includes automatic frameworks to synthesize approximate circuit blocks, a model for programmable approximate processors that explicitly codifies the notion of quality into the HW/SW interface, and finally software techniques to systematically identify resilient computations within an application and to apply approximate computing to achieve a favorable quality-efficiency tradeoff. I will conclude with an overview of the other research directions that I am exploring to address the efficiency gap viz. computing with spintronics, and heterogeneous many-core accelerators for emerging workloads.

Biography: Swagath Venkataramani is a 5-year PhD student in the School of Electrical and Computer Engineering, Purdue University. His research interests include, Approximate Computing, Computing with Spintronic Devices, Heterogeneous Parallel Architectures, and Computational Imaging. His dissertation research was awarded the Intel PhD fellowship in computing leadership and Purdue Bilsland Dissertation fellowship. It has also been featured in MIT Technology Review, Slashdot, Physics Today, and NSF News from the Field. Swagath graduated with a Bachelors degree in Electrical and Electronics Engineering from College of Engineering, Guindy, Anna University, India as the university gold medalist. He has worked with the Exa-scale Computing Group at Intel as part of the US DOE’s FastForward Program, and with the Sensing and Energy Research Group at Microsoft Research.

Host: Prof. Alice C. Parker

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Annie Yu