Author(s): Bheemala Vamsi

Abstract: Training datasets are available in open source and these training datasets are used by the researchers in various discipline. To implement their research in those datasets are to simulate their results in various measures like comparative study, accuracy and prediction. But the reliability of these datasets is questionable and they are subject to various attack. Machine learning algorithm results are strongly manipulated when the learning algorithms evaluated the causative attack, evasion attack and membership inference attacked training datasets. So, the training datasets of image datasets, Natural language datasets, and pattern recognition datasets are to be protecting from various attacks. In this paper concentrate the overview of training datasets attacks and provide the security to the real training dataset.

DOI: 10.33545/27076636.2020.v1.i2a.11

Pages: 13-19 | Views: 707 | Downloads: 245

Download Full Article: Click Here