ATELIER. Self-Labelling dynamic Loss function training data

This dataset contains all the statistics collected during the training of multiple Neural Networks, which uses the following architecture technologies: Siamese Neural Networks, Dynamic loss function, modified during the training through Reinforcement learning and a 'Curriculum-Learning' cycle in between training cycles. The goal of the Models is to correctly identify anomalies in the QoE of real-time videos through the network KPIs. The dataset contains 3 type of experiment with different levels of model complexity. Please refer to the associated github repository and the published paper for a detailed description of both the dataset and the infrastructure that generated the dataset.Journal paper title: "ATELIER: Service Tailored and Limited-Trust Network Analytics Using Cooperative Learning"