EagerNet: Early Predictions of Neural Networks for Computationally Efficient Intrusion Detection

摘要: Fully Connected Neural Networks (FCNNs) have been the core of most state-of-the-art Machine Learning (ML) applications in recent years and also widely used for Intrusion Detection Systems (IDSs). Experimental results from last show that generally deeper neural networks with more layers perform better than shallow models. Nonetheless, growing number layers, obtaining fast predictions less resources has become a difficult task despite use special hardware such as GPUs. We propose new architecture to detect network attacks minimal resources. The is able deal either binary or multiclass classification problems trades prediction speed accuracy network. evaluate our proposal two different intrusion detection datasets. Results suggest it possible obtain comparable accuracies simple FCNNs without evaluating all majority samples, thus early saving energy computational efforts.