Prof. Kannan Ramchandran (UC Berkeley)

Seven decades after Claude Shannon's groundbreaking work, codes are now an indispensable part of modern communications and storage systems. But do they have a role in the modern era of big data deluge and machine learning? How can codes help address the challenge of scale in speeding up computation, learning and inference by exploiting the underlying structure such as sparsity? In this talk, we will explore how coding theory can offer an interesting playground for diverse problem settings, illustrated through the lens of sparse-graph codes. We will show how this forms the core of a unified architecture featuring a divide-and-conquer strategy based on fast peeling-based decoding. We will highlight our approach to computational tasks such as massive-scale sparse Fourier and Walsh decompositions, sparse polynomial learning, compressed sensing, group testing, and learning mixtures of sparse linear regressions. Applications include fast acquisition for magnetic resonance imaging (MRI), graph sketching, fast neighbor discovery in the IoT, and spectrum sensing for cognitive radio. We will also survey how codes can speed up distributed machine learning in today's cloud computing platforms by rendering them robust to system noise in the form of straggling compute nodes, with a particular focus on server-less systems such as Amazon's AWS Lambda.

Kannan Ramchandran is a Professor of Electrical Engineering and Computer Science at UC Berkeley, where he has been since 1999. Prior to that, he was on the faculty of the University of Illinois at Urbana-Champaign from 1993 to 1999. Prof. Ramchandran is a recipient of the IEEE Kobayashi Computers and Communications Award for his pioneering contributions to theory and practice of distributed storage codes and distributed compression. He is a Fellow of the IEEE, has published extensively, and holds more than a dozen patents. He has received several awards for his research and teaching including an IEEE Information Theory Society and Communication Society Joint Best Paper award for 2012, an IEEE Communication Society Data Storage Best Paper award in 2010, two Best Paper awards from the IEEE Signal Processing Society in 1993 and 1999, an Okawa Foundation Award for outstanding research at Berkeley in 2001, a Hank Magnusky Scholar award at Illinois in 1998, and an EECS Departmental Outstanding Teaching award at Berkeley in 2009. His current research interests are at the intersection of coding theory, statistical signal processing, and distributed machine learning.