Abstract: Cancer diagnosis is usually an arduous task for medicine, specially when it comes to pulmonary cancer, which is one of the most deadly and hard to treat types of cancer. Early detection of pulmonary cancerous nodules drastically increases surviving chances, but also makes it an even harder problem to solve, as it mostly depends on a visual inspection of tomography scans. To help improving this detection and surviving rates, engineers and scientist have been developing computer-aided diagnosis techniques, as the one presented in this paper. Here, we use computational intelligence to propose a new approach towards solving the problem of detecting pulmonary carcinogenic nodules in computerized tomography scans. The technology applied consists in using Deep Learning and Swarm Intelligence to develop a novel nodule detection and classification model. Seven different Swarm Intelligence algorithms and Convolutional Neural Networks for biomedical image segmentation are used to detect and classify cancerous pulmonary nodules in the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI). The aim of this work is to train Convolutional Neural Networks using swarm intelligence techniques and demonstrate that this approach is more efficient than the classic training with Back-propagation and Gradient Descent. It improves the average accuracy from 93% to 94%, precision from 92% to 94%, sensitivity from 91% to 93% and specificity from 97% to 98%, which constitute a relevant improvement regarding the statistical T-test.

Abstract: Network anomaly detection has caused widespread concern among researchers and the industry. Existing work mainly focuses on applying machine learning techniques to detect network anomalies. The ability to exploit the potential relationships of communication patterns in network traffic has been the focus of many existing studies. Graph kernels provide a powerful means for representing complex interactions between entities, while deep neural networks break through new foundations for the reason that data representation in the hidden layer is formed by specific tasks and is thus customized for network anomaly detection. However, deep neural networks cannot learn communication patterns among network traffic directly. At the same time, deep neural networks require a large amount of training data and are computationally expensive, especially when considering the entire network flows. For these reasons, we employ a novel method to marry graph kernels to deep neural networks, which exploits the relationship expressiveness among network flows and combines ability of neural networks to mine hidden layers and enhances the learning effectiveness when a limited number of training examples are available. We evaluate the proposed method on two real-world datasets which contains low-intensity network attacks and experimental results reveal that our model achieves significant improvements in accuracies over existing network anomaly detection tasks.

Abstract: We explore the idea of integrating machine learning with simulations to enhance the performance of the simulation and improve its usability for research and education. The idea is illustrated using hybrid openMP/MPI parallelized molecular dynamics simulations designed to extract the distribution of ions in nanoconfinement. We find that an artificial neural network based regression model successfully learns the desired features associated with the output ionic density profiles and rapidly generates predictions that are in excellent agreement with the results from explicit molecular dynamics simulations. The results demonstrate that the performance gains of parallel computing can be further enhanced by using machine learning.

Abstract: Cross-modal information retrieval (CMIR) enables users to search for semantically relevant data of various modalities from a given query of one modality. The predominant challenge is to alleviate the "heterogeneous gap" between different modalities. For text-image retrieval, the typical solution is to project text features and image features into a common semantic space and measure the cross-modal similarity. However, semantically relevant data from different modalities usually contains imbalanced information. Aligning all the modalities in the same space will weaken modal-specific semantics and introduce unexpected noise. In this paper, we propose a novel CMIR framework based on multi-modal feature fusion. In this framework, the cross-modal similarity is measured by directly analyzing the fine-grained correlations between the text features and image features without common semantic space learning. Specifically, we preliminarily construct a cross-modal feature matrix to fuse the original visual and textural features. Then the 2D-convolutional networks are proposed to reason about inner-group relationships among features across modalities, resulting in fine-grained text-image representations. The cross-modal similarity is measured by a multi-layer perception based on the fused feature representations. We conduct extensive experiments on two representative CMIR datasets, i.e. English Wikipedia and TVGraz. Experimental results indicate that our model outperforms state-of-the-art methods significantly. Meanwhile, the proposed cross-modal feature fusion approach is more effective in the CMIR tasks compared with other feature fusion approaches.

Abstract: The deep neural networks (DNNs), although highly accurate, are vulnerable to adversarial attacks. A slight perturbation applied to a sample may lead to misprediction of the DNN, even it is imperceptible to humans. This defect makes the DNN lack of robustness to malicious perturbations, and thus limits their usage in many safety-critical systems. To this end, we present DunDi, a metric learning based classication model, to provide the ability to defend adversarial attacks. The key idea behind DunDi is a metric learning model which is able to pull samples of the same label together meanwhile pushing samples of dierent labels away. Consequently, the distance between samples and model's boundary can be enlarged accordingly, so that signicant perturbations are required to fool the model. Then, based on the distance comparison, we propose a two-step classication algorithm that performs eciently for multi-class classication. DunDi can not only build and train a new customized model but also support the incorporation of the available pre-trained neural network models to take full advantage of their capabilities. The results show that DunDi is able to defend 94.39% and 88.91% of adversarial samples generated by four state-of-the-art adversarial attacks on the MNIST dataset and CIFAR-10 dataset, without hurting classication accuracy.