Hadi Esmaeilzadeh transferred from UT Austin to the University of Washington several years ago, when his advisor Doug Burger joined Microsoft Research.
Since his arrival at UW, Hadi has had an extraordinary string of three IEEE Micro “Top Picks” papers – roughly ten papers selected annually as the very best to have appeared in the various computer architecture conferences during the year, and re-printed in a special issue of IEEE Micro.
Hadi’s third “Top Picks” paper has just appeared: “Neural Acceleration for General-Purpose Approximate Programs,” co-authored with fellow UW CSE graduate student Adrian Sampson, and with Hadi’s co-advisors, Burger and UW CSE’s Luis Ceze. The paper proposes an approximate algorithmic transformation and a new class of accelerators, Neural Processing Units (NPUs). NPUs leverage our approximate algorithmic transformation that converts regions of code from a Von Neumann model to a neural model and achieve an average 2.3x speedup and 3.0x energy savings for general-purpose approximate programs. This new class of accelerators show that significant performance and efficiency gains are possible when the abstraction of full accuracy is relaxed in general-purpose computing.
Hadi is completing his Ph.D. and will join the faculty at the Georgia Institute of Technology in the fall as the first holder of the Catherine M. and James E. Allchin Early Career Professorship.