Search

Deep neural networks (DNNs) have achieved astonishing results on a variety of supervised learning tasks owing to a large scale of well-labeled training data. However, as recent researches have pointed out, the generalization performance of DNNs is likely to sharply deteriorate when training data contains label noise. In order to address this problem, a novel loss function is proposed to guide DNNs to pay more attention to clean samples via adaptively weighing the traditional cross-entropy loss. Under the guidance of this loss function, a cross-training strategy is designed by leveraging two synergic DNN models, each of which plays the roles of both updating its own parameters and generating curriculums for the other one. In addition, this paper further proposes an online data filtration mechanism and integrates it into the final cross-training framework, which simultaneously optimizes DNN models and filters out noisy samples. The proposed approach is evaluated through a great deal of experiments on several benchmark datasets with man-made or real-world label noise, and the results have demonstrated its robustness to different noise types and noise scales.

In this article, we focus on the usage of MQTT (Message Queuing Telemetry Transport) within Connected Vehicles (CVs). Indeed, in the original version of MQTT protocol, the broker is responsible “only” for sending received data to subscribers; abstracting then the underlying mechanism of data exchange. However, within CVs context, subscribers (i.e., the processing infrastructure) may be overloaded with irrelevant data, in particular when the requirement is real or near real-time processing. To overcome this issue, we propose MQTT-CV; a new variant of MQTT protocol, in which the broker is able to perform local processing in order to reduce the workload at the infrastructure; i.e., filtering data before sending them. In this article, we first validate formally the correctness of MQTT-CV protocol (i.e., the three components of the proposed protocol are correctly interacting), through the use of Promela language and its system verification tool; the model checker SPIN. Secondly, using real-world data provided by our car manufacturer partner, we have conducted real implementation and experiments. The obtained results show the effectiveness of our approach in term of data workload reduction at the processing infrastructure. The mean improvement, besides the fact that it is dependent of the target application, was in general about 10 times less in comparison to native MQTT protocol. [ABSTRACT FROM AUTHOR]