2018 Lecture Series with Gauri Joshi, Carnegie Mellon University
“Distributed Storage” – video
Erasure codes, originally designed to provide reliability against noise on communication channels, are now widely used in distributed storage systems. Distributed storage systems require the codes to have new properties such as low-cost repair of failed nodes, high service capacity, and fast content access. We will discuss new classes of codes designed to achieve each of these properties.
“Coded Computing” – video
Large-scale distribute computing frameworks such as MapReduce and Spark employ massive parallelization of jobs. While parallelism drastically reduces computation time, slow or straggling nodes can become a bottleneck in job completion. In this talk, we will study erasure codes for speeding up various types of distributed computation such as matrix-vector and matrix-matrix multiplications, convolutions and gradient descent.
“Rateless Codes” – video
Rateless fountain codes are a class of erasure codes where the source generates an unlimited stream of symbols until the data is recovered at the receiver. This rateless property makes them well-suited for many applications including multicast communication, distributed storage, and streaming. In this talk, we will cover the fundamental concepts of rateless fountain codes and their applications.