The Classification of Multiple-domain Samples based on Multiple Deep Learning Models

What is it about?

This paper proposed a classification method that can classify the samples of multiple domains based on the cooperation of multiple models. On a sample, our method firstly predicts the domain that may contains this sample based on the output of trained models. Then, our method uses the models that are trained on this domain to classify the label of this sample. Our method can reduce the difficulty of training high performance model while can ensure high accuracy on the samples of multiple domains. Experiments show that our method achieved higher accuracy than the existing methods.

Featured Image

Why is it important?

This paper proposed a multi-domain classification method based on probabilities of labels. These probabilities are the output of trained models. On each domain, we train some existing models on the training set of this domain. On other domains, we do not transfer or retraining these models, which benefits the scalability of a deep learning model system. Then our system can classify the samples of multiple domains. Firstly, on a sample, we use the outputs of all models to predict the domain that may contain this sample. Secondly, we use the trained models of that domain to predict the label of this sample.