Kavitha Ganesan: Machine Learning Data Detection Poisoning Attacks Using Resource Schemes Multi-Linear Regression. Neural Parallel and Scientific Computations 28 (2020), No. 2, 73-82.

https://doi.org/10.46719/npsc20202821

ABSTRACT.
Machine learning methods become more and more popular. The purpose of using these methods is to become more and more popular network security components such as firewalls and anti-virus software such as machine learning methods expected to rise. Data machine learning systems provided by well-trained users can be vulnerable to attacks, poisoning data where malicious users inject fake training data, and damage the learning model. Data poisoning attacks can damage the integrity of the machine learning model by introducing malicious training models that affect results during testing. Distributed machine learning (DML) and Semi-DML is training that can be realized from a large database when any node is able to work out accurate results at an acceptable time. Compared to this inevitably diverse environment the attack will still expose potential targets. In this proposed method, we introduced the method for data detection poisoning, Data Poison Detection Program, Resource Schemes Multi-Linear Regression (RSMLR) to provide better learning protection and assistance from central sources. Proper allocation of resources in RSMLR can reduce resource waste. The application of modifying the data poison detection program can extend the system even more dynamically according to the environment and attack intensity. In addition, many of the components will increase the resource consumption of the system due to training.

Key words: Machine learning, Data detection, poisoning scheme, Resource Schemes Multi-Linear Regression (RSMLR), Distributed machine learning (DML) and Semi-DML.