Krishna Palem

Kenneth and Audrey Kennedy Professor of Computing

Krishna Palem

Career & Awards

For Dr. Palem’s bio, download the PDF.

Career Narrative – Inexact Computing

While I started out my career as a member of IBM T. J. Watson Research Center’s technical staff in 1986, the interesting part of how my work led to inexact or approximate computing started in 2001. During this time, I started actively investigating the question of energy and power consumption especially in portable embedded systems such as cell phones which are naturally power limited. My training and early career involved working on foundational questions in computer science. As part of this, I worked on techniques used in probabilistic or randomized algorithms which could achieve efficiencies in their speed by trading the probability of the answer being correct by a small amount. Since I was interested not in abstract running time but rather in the physical energy of computing systems, I started thinking of this problem in the context of a construct that Richard Feynman used in one of his Lectures on Computation given at Caltech. In these lectures, Feynman determined through an entropy argument that the minimum amount of energy needed to compute a bit is kT ln (2) joules; the original articulation of this insight is attributed to von Neumann.

Principled approach to inexactness: My first three papers that eventually led to work by us and others in inexact computing connected Feynman’s “device’’ with the concept of inexact computing. At the most fundamental level, I showed that a single bit of information has two attributes: quality (or correctness) which randomization traditionally embraced, and an associated cost. Specifically, I was able to show that the minimum energy to compute a bit can be lower than the limit from Feynman’s lectures based on the following central thesis: the accuracy of computations can be relaxed or traded for energy savings. My insight was to show that the probability with which bits are computed correctly can be traded for energy consumption in a principled manner and consequently, the lower the probability of being correct, the lower the energy consumed.  As far as can be determined, this was the first characterization of this relationship. This led to several additional papers which we developed through my group, housed at the Center for Research in Embedded Systems and Technology at Georgia Tech. This work provided validations of my original thesis using CMOS semiconductors, the ubiquitous foundation through which modern computer chips are realized. Much of the development that followed along this somewhat unorthodox approach of garnering significant gains in energy, speed and the physical area occupied by computing devices by trading away the quality of the computation starting with individual bits, was pursued at Georgia Tech, with support from Hewlett Packard and federal funding sources.

Over the  past decade, I became increasingly interested in applying inexactness to problems of societal value. My first foray was into education where I conceived and announced the I-Slate in 2008. The I-Slate involved combining the concept of low-energy embedded computing with highly usable user environments to provide very low cost self-tutoring devices in economic and electric resource challenged environments, as well as through the concept of sensoptimization to demonstrate the value of inexactness in the context of hearing aids.

Inexact Computing and Embedded Computing Awards and Recognition

1. Guggenheim Fellowship, 2015.

2. Keynote at two major AC<M international symposia in 2014

  • ACM Symposium on Programming Language Design and Implementation (PLDI), Edinburgh, UK.
  • ACM  23rd  International  Symposium  on  High-performance  Parallel  and Distributed Computing (HPDC), Vancouver, BC.

3. Best Paper Award at ACM International Conference on Computer Frontiers 2012.

4. Best Paper Award at the International Conference on Architecture of Computing Systems (ARCS) 2050.

5. Ranked No.2 by Forbes India in March 2012 among a list of 18 people of Indian origin world-wide who are “changing the world” and “…finest minds of Indian origin….”.

6. Fellow American Association for the Advancement of Science (AAAS),2012.

7. Feynman Prize in Nanotechnology (Theory), Finalist, 2011.

8. Fellow of the American Association for the Advancement of Science, 2011.

9. I-Slate, one of Seven World-Changing Technologies, featured as part of IEEE’s 125th Anniversary, 2009.

10. IEEE W. Wallace McDowell Award, 2008.

  • This is the highest award for technical achievement given solely by the IEEE Computer Society and likened in the public domain to the ACM Turing Award and referred to sometimes as the “IT Nobel’’ (http://en.wikipedia.org/wiki/W._Wallace_McDowell_Award)

11. Probabilistic Chips (PCMOS), one of 10 technologies that we think are most likely to change the way we live, Technology Review published by MIT, 2008

12. PCMOS, recognized by Technology Review published by MIT as one of the “10 technologies that we think are most likely not to change the way we live,” 2008.

13. Moore Distinguished Faculty Fellow, Caltech, 2006-07.

14. Canon Visiting Professor at Nanyang Technological University, Singapore.

15. Fellow Association for Computing Machinery (ACM), 2006.

16. Invited Professor, École Normale Supérieure, Paris, France, 2004–05.

17. Fellow Institute of Electrical and Electronics Engineers (IEEE),  2004.

18. One of the four nominees for outstanding technologies of 2002, Analysts Choice Awards

  • For DVAITA, a technology created through our research and commercialized by Proceler Inc. of which I was the founding CTO.

19. Schonbrunn Fellow, The Hebrew University, Jerusalem, Israel, 2000.

20. Teaching Excellence, The Hebrew University, Jerusalem, Israel, 1999.

21. Three  dissertation awards

  • S. Talla, Janet Fabri Award for Outstanding Doctoral Thesis, “Adaptive Explicitly Parallel Instruction Computing” May 2001, New York University.
  • Leung and S. Talla, Harold Grad Prize, April 1998 and Dean’s Dissertation Fellowship, September 1999, respectively, New York University.

22. External Recognition Award, IBM Research Division, 1994.

Over the past five years, several groups around the world have started working on interpretations of inexactness, and scores of papers have been published in a peer-reviewed context—the topic is now popularly referred to as inexact or approximate computing.

Inexactness for supercomputing, weather and climate modeling

In 2007, I moved to Rice University as the first Ken and Audrey Kennedy professor, and my own interest shifted towards applying inexact computing to benefit yet another domain of societal value—climate modeling and weather prediction within the broader milieu of supercomputing. Professor Tim Palmer, FRS of Oxford University, a pioneer in climate science and our collaborator on this topic, notes the many benefits to models with such improved resolution.  Notably, Professor Palmer has concluded that climate simulations with an order of magnitude better resolution than those available today are needed to reduce uncertainties in estimates of global warming. Through work with Professor Palmer’s group, our results have shown that inexact computing through could be a definite direction for improving resolution while preserving existing energy profiles. Two noteworthy articles in the public domain or press that have articulated this approach are:

Pioneering role within embedded computing within CS in the 1990s

My foray into inexact computing stemmed from my interest in applying rigorous computer science principles to tackle challenges in emerging domains and at the boundary. Historically, computer science helped create wonderful abstractions—high level programming languages and operating systems for example—which enabled users to develop applications with great ease and facility. In 1995, I set up a laboratory, considered to be among the earliest in computer science focused on embedded computing, at the Courant Institute of Mathematical Sciences where I was a tenured faculty member. Over the course of the next six years, the laboratory and its research blossomed into having very broad impact ranging including creating new conferences,

  • The now thriving ACM-IEEE Symposium on Compilers, Architecture and Synthesis for Embedded Computing
    • o one of the three flagship conferences of the Embedded Systems Week.

Public policy

At Rice university, I am also a scholar in the non-partisan Baker institute of public policy studying the policy implications of energy as it relates to the information technology sector. Also, in 2008, I founded a joint institute between Rice University and Nanyang Technological University in Singapore, which I directed for five years. The goals of this institute were to innovate and discover sustainable technologies with focus in energy consumption