We found an easy and quick post-learning method named "Icing on the Cake" to enhance a classification performance in deep learning. The method is that we train only the final classifier again after an ordinary training is done.
Classifier Extract Feature maps
Re-trained Classifier ClassifierRe-train only the classifier
Put backFigure 1: The sketch of the proposed method.("Icing on the Cake"). Left: Train a deep neural network as usual. Center: Extract features of input data as estimation from the layer before the final classifier, and then train the final classifier again by the extracted features. Right: Put the re-trained classifier back to the network.
The Universal Communication Research Institute (UCRI), NICT conducts research and development on universal communication technologies: multi-lingual machine translation, spoken dialogue, information analysis and ultrarealistic interaction technologies, through which people can truly interconnect, anytime, anywhere, about any topic, and by any method, transcending the boundaries of language, culture, ability and distance. To enhance the universal communication technology, we are trying to develop a large-scale information infrastructure which collects and stores diverse information including huge volumes of web pages from the networks. The one of most important key technologies to realize a large-scaled information infrastructure is a distributed in memory database system. In this paper, we introduced a large-scale information infrastructure, mainly explaining a distributed in-memory database system "okuyama" which is a key technology on our project. We examined the I/O performance of the in memory storage, which verified whether if "okuyama" meet requirements for the infrastructure. Furthermore we give a blueprint of cluster systems on which the infrastructure will be constructed.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.