Coded distributed computing for speeding up massively important computations

Abstract: The main goal is to address faults and straggling in distributed computing systems through coding. We are focusing on applications such as machine learning, protein folding, neural data analysis for clinical diagnoses, etc.

Our lab has led the formalization of use of error-correcting codes to address these challenges. Our papers that have appeared in NIPS'16 and IEEE Trans Info Theory '17 establish the utility of such approaches. The student will apply these ideas to practical distributed systems such as Amazon EC2. A solid programming background, and good mathematical background is required. The student will interact with 2-3 Ph.D. students from our lab.