EagerNet: Early Predictions of Neural Networks for Computationally Efficient Intrusion Detection
This work addresses the need for efficient intrusion detection systems in network security, though it is incremental as it builds on existing neural network methods.
The paper tackles the problem of computationally expensive intrusion detection with deep neural networks by proposing EagerNet, an architecture that enables early predictions to save resources while maintaining comparable accuracy to simple fully connected networks on two datasets.
Fully Connected Neural Networks (FCNNs) have been the core of most state-of-the-art Machine Learning (ML) applications in recent years and also have been widely used for Intrusion Detection Systems (IDSs). Experimental results from the last years show that generally deeper neural networks with more layers perform better than shallow models. Nonetheless, with the growing number of layers, obtaining fast predictions with less resources has become a difficult task despite the use of special hardware such as GPUs. We propose a new architecture to detect network attacks with minimal resources. The architecture is able to deal with either binary or multiclass classification problems and trades prediction speed for the accuracy of the network. We evaluate our proposal with two different network intrusion detection datasets. Results suggest that it is possible to obtain comparable accuracies to simple FCNNs without evaluating all layers for the majority of samples, thus obtaining early predictions and saving energy and computational efforts.