ARCS Scholar Tyler Sypherd from Arizona State University was found being “properly improper” as he presented his research at the International Conference on Machine Learning (ICML) - the leading international academic conference in machine learning, hosted in Baltimore this past July.
Sypherd’s paper titled “Being Properly Improper” explores how to make machine learning models more robust under data corruption by using special loss functions to shape the learning process. The paper was accepted for presentation in a short talk and a poster at the ICML which only accepts 20 percent of submitted papers.
“Being Properly Improper” was co-authored with ASU Professor Lalitha Sankar (and Sypherd’s PhD advisor) and Google researcher Richard Nock.
“Often in practice when you are training a machine learning algorithm, the data is noisy, corrupted [and] imbalanced,” Sypherd explained. “We are arguing that one should consider a different type of loss function - a different type of machine teacher in scenarios where you know there is corrupted data.”
Sypherd’s paper was included in one of ICML’s spotlight sessions, the conference’s Healthcare AI and COVID-19 workshop, and Interpretable Machine Learning and Healthcare workshop.