A Unified Framework for Analyzing and Detecting Malicious Examples of DNN Models