Workshop 15: MPP Parallel Programming Models - Special Edition Machine Learning Performance and Security

Recent publications show that, since it is clear that machine learning has become ubiquitous, it is necessary to produce research to increase currently ML algorithms’ performance, power consumption and, most importantly as of recently, security. The performance issues are quite clear to understand and have been addressed by multiple researchers either with new algorithms for model compression/parallelization or with new specific hardware. Roughly speaking, it is possible to categorize the security threats against ML models in privacy leakage and model evasion (”fooling” the system into making wrong decisions). This edition of the Workshop on Parallel Programming Models (MPP) is focused on addressing theses issues which are of the utmost importance for the Machine Learning community at this point.