In this talk, I will present a unifying optimization framework to make massive data analysis tasks less sensitive to corrupted and missing data by exploiting domain specific knowledge and prior information about structure.