Search

Deep learning is a powerful tool for domain adaptation by learning robust high‐level domain invariant representations. Recently, adversarial domain adaptation models are applied to learn representations with adversarial training manners in feature space. However, existing models often ignore the generation process for domain adaptation. To tackle this problem, deep cycle autoencoder (DCA) is proposed that integrates a generation procedure into the adversarial adaptation methods. The proposed DCA consists of four parts, a shared encoder, two separated decoders, a discriminator and a linear classifier. With the labelled source images and unlabelled target images as inputs, the encoder extracts high‐level representations for both source and target domains, and the two decoders reconstruct the inputs with the latent representations separately. The shared encoder is pitted against the discriminator; the encoder tries to confuse the discriminator while discriminator aims at distinguishing which domain the latent representations come from. DCA adopts both adversarial loss and maximum mean discrepancy loss in the latent space for distribution alignment. The classifier is trained with both the source original and reconstructed image representations. Extensive experimental results have demonstrated the effectiveness and the reliability of the proposed methods.

Abstract Background In recent years, research in artificial neural networks has resurged, now under the deep-learning umbrella, and grown extremely popular. Recently reported success of DL techniques in crowd-sourced QSAR and predictive toxicology competitions has showcased these methods as powerful tools in drug-discovery and toxicology research. The aim of this work was dual, first large number of hyper-parameter configurations were explored to investigate how they affect the performance of DNNs and could act as starting points when tuning DNNs and second their performance was compared to popular methods widely employed in the field of cheminformatics namely Naïve Bayes, k-nearest neighbor, random forest and support vector machines. Moreover, robustness of machine learning methods to different levels of artificially introduced noise was assessed. The open-source Caffe deep-learning framework and modern NVidia GPU units were utilized to carry out this study, allowing large number of DNN configurations to be explored. Results We show that feed-forward deep neural networks are capable of achieving strong classification performance and outperform shallow methods across diverse activity classes when optimized. Hyper-parameters that were found to play critical role are the activation function, dropout regularization, number hidden layers and number of neurons. When compared to the rest methods, tuned DNNs were found to statistically outperform, with p value

Many real-world applications, such as bioinformatics, data mining, pattern recognition, and social network analysis, benefit from efficient solutions for the graph similarity search problem. Existing methods have limited scalability when they handle the large graph databases, for example, those with millions or billions of graphs that cannot fit in main memory. In this paper, we study the problem of graph similarity search under the graph edit distance constraint in external memory. We present an efficient framework for arbitrary q-gram-based representations of a graph. Specifically, we propose a q-gram matrix index stored in hybrid layout in external memory to achieve efficient query processing, by converting the q-gram counting filter into a sparse matrix-vector multiplication problem. Furthermore, we also boost the query performance by transforming the global filter to a 2-D query rectangle, which allows us to perform a query in a reduced region, significantly reducing the number of query I/Os in practice. Extensive experiments on real data sets confirm that 1) our method can compete with the state-of-the-art in-memory methods in index size and filtering ability, and outperform them on scalability of coping with the PubChem data set including 25 million chemical structure graphs and 2) compared with the popular q-gram-based external inverted index, our external index structure needs much fewer number of query I/Os on the PubChem data set.