Konstantinos Konstantinidis, Ph.D. student with Iowa State University‘s Department of Electrical and Computer Engineering (ECpE), recently won first prize in the Student Poster Award contest at the Midwest Machine Learning Symposium (MMLS) 2019. Konstantinidis, who performed his research jointly with ECpE Professor Aditya Ramamoorthy, won for his poster titled, “Speeding up Distributed Computing through Coding.”
Konstantinidis’ poster focuses on distributed protocols, such as MapReduce, used for big data analytics on clusters. Specifically, it illustrates the magnitude of the communication overhead of the Shuffle phase of these algorithms and suggests a method to alleviate it. The method describes a relationship between these protocols and combinatorial structures known as resolvable designs. These combinatorial structures allow for a judicious assignment of tasks to servers and manage to reduce the communication time by a significant factor compared to the baseline implementation and prior techniques. The method is extended to a class of functions used in machine learning called aggregate functions, and experimental results obtained on Amazon EC2 clusters are presented.
In the MMLS 2019 poster competition, out of the 80 accepted posters, eight were chosen as finalists. Posters were judged based on a three-minute spotlight talk, abstract and the discussion at the poster session. The award comes with a $1,000 cash prize.