Regularized Chained Deep Neural Network Classifier for Multiple Annotators

The increasing popularity of crowdsourcing platforms, i.e., Amazon Mechanical Turk, changes how datasets for supervised learning are built. In these cases, instead having labeled by one source (which is supposed to be an expert who provided the absolute gold standard), databases holding multiple annotators provided. However, most state-of-the-art methods devoted from experts assume that labeler’s behavior homogeneous across input feature space. Besides, independence constraints imposed on annotators’ outputs. This paper presents a regularized chained deep neural network deal with classification tasks annotators. introduced method, termed RCDNN, jointly predicts ground truth label and performance space samples. turn, RCDNN codes interdependencies among analyzing layers’ weights includes l1, l2, Monte-Carlo Dropout-based regularizers over-fitting issue in models. Obtained results (using both simulated real-world annotators) demonstrate can multi-labelers scenarios tasks, defeating techniques.