When no one node can produce accurate results in a reasonable amount of time, distributed machine learning (DML) can be used to train enormous datasets. However, in comparison to a non-distributed environment, this will necessarily expose more possible targets to attackers. We divide DML into two categories in this paper: basic-DML and semi-DML. The center server in basic-DML assigns learning tasks to dispersed devices and combines their results.
In semi-DML, the center server allocates additional resources to dataset learning in addition to its basic-DML responsibilities. To begin, we propose a novel data poison detection approach for basic-DML that uses a cross-learning mechanism to locate the poisoned data. We build a mathematical model to discover the ideal number of training loops based on proving that the suggested cross-learning mechanism would create training loops.
Then, for semi-DML, we provide an improved data poison detection approach that uses the central resource to give greater learning protection. An optimal resource allocation technique is designed to make the most of the system’s resources. In the basic-DML scenario, simulation results suggest that the proposed strategy can enhance the accuracy of the final model by up to 20% for support vector machines and 60% for logistic regression. Furthermore, in a semi-DML scenario, a better data poison detection technology combined with optimum resource allocation can save wasted resources by 20-100 percent.
The existing model of the system:-
?Supervised approach algorithms
Accuracy is not to a level
Proposed model of the system:-
SVM classifier,? DML, semi DML? scenarios have been considered
Accurate results within less time.
Operating system-windows 7,8,10
Anaconda navigator,jupyter notebook,python language (IDLE)