Abstract: Critical systems are increasingly being integrated with machine learning (ML) models, which exposes them to a range of adversarial attacks.The vulnerability of machine learning systems to hostile attacks has drawn a lot of attention in recent years. When harmful input is added to the training set, it can lead to poison attacks, which can seriously impair model performance and threaten system security. Poison attacks pose a serious risk since they involve the injection of malicious data into the training set by adversaries, which influences the model’s performance during inference. It’s necessary to identify these poison attacks in order to preserve the reliability and security of machine learning systems. A novel method based on transfer learning is proposed to identify poisoning attacks in machine learning systems.The methodology for generating poison data is initially created and later implemented using transfer learning techniques. Here, the poisonous data is detected using the pre-trained VGG16 model. This method can also be used in distributed Machine learning systems with scattered data and computation across several nodes. Benchmark datasets are used to evaluate this strategy in order to prove the effectiveness of proposed method. Some real-time applications, advantages, limitations and future work are also discussed here.

Archa A T and K. Kartheeban. “SecureTransfer: A Transfer Learning Based Poison Attack Detection in ML Systems”. International Journal of Advanced Computer Science and Applications (IJACSA) 15.7 (2024). http://dx.doi.org/10.14569/IJACSA.2024.01507141

@article{T2024,
title = {SecureTransfer: A Transfer Learning Based Poison Attack Detection in ML Systems},
journal = {International Journal of Advanced Computer Science and Applications},
doi = {10.14569/IJACSA.2024.01507141},
url = {http://dx.doi.org/10.14569/IJACSA.2024.01507141},
year = {2024},
publisher = {The Science and Information Organization},
volume = {15},
number = {7},
author = {Archa A T and K. Kartheeban}
}