A Case Study of Inductive Transfer Learning in Intrusion Detection System

The Chapter Five presents a base for further research and discussions on how an IDS model implemented for some “A” attacks can be utilized using Inductive Transfer Learning to learn about other “B” attacks considering feature space for both “A” and “B” attacks is same. To evaluate Inductive Transfer Learning in the field of Information Security. For Information Security, Network Intrusion Detection System is considered. Using IDS, organizations protect their network against attacks like Denial of Service, PortScan, Heartbleed, and other cyberattacks. As a solution, we are proposing a fine-tuning-based transfer learning approach. The proposed inductive learning method can assist in developing a new IDS model from pre-developed IDS for other attacks. Using this approach, the problem of implementing a new IDS from scratch can be avoided because implementing a new IDS from scratch can be difficult sometimes. For instance, if any Deep Learning algorithm such as DNN is considered, then to implement a new DNN from scratch, several things need to be considered like deciding the number of hidden layers, the number of nodes in each hidden layer, optimization function, learning rates, and other important parameters for DNN model. By applying Inductive TL, the questions like deciding the number of hidden layers or number of nodes in hidden layers can be answered by transferring a pre-developed source model to learn about the target task.