Spatial co-location pattern mining (SCPM) aims to discover subsets of spatial features frequently located together in proximate areas. Previous studies of SCPM solely concern the inter-features association of a pattern, but neglect the interesting intra-feature behavior. In this paper, we propose the task of spatial co-location pattern mining with coupling relation consideration (SCPM-CR) to capture complex relations embedded in a co-location. Specifically, InterPCI measure is designed to capture the inter-features coupling by considering the comprehensive interaction of objects for the features in a pattern, and luckily it possesses the anti-monotone property. Another measure, IntraCAI, is proposed to capture the congregating behavior of intra-feature objects under the restriction of a co-location. A general framework is designed for SCPM-CR task and experiments show that a large fraction of computation time is devoted to identifying the participating objects of a candidate pattern. To tackle this calculation bottleneck, a novel candidate-and-search algorithm is suggested, CS-HBS, equipped with heuristic backtracking search. Extensive experiments are conducted on real and synthetic datasets to demonstrate the superiority of SCPM-CR compared with traditional SCPM methods, and also to validate the efficiency and scalability of CS-HBS. Experimental results show that CS-HBS outperforms the baselines by several times or even orders of magnitude.
2 December 2019 in Distributed and Parallel Databases
Deep neural networks (DNNs) are sensitive to adversarial examples which are generated by corrupting benign examples with imperceptible perturbations, or have significant changes but can still achieve original prediction results. The latter case is termed as the Type I adversarial example which, however, has limited attention in the literature. In this paper, we introduce two methods, termed HRG and GAG, to generate Type I adversarial examples and attempt to apply them to the privacy-preserving Machine Learning as a Service (MLaaS). Existing methods for the privacy-preserving MLaaS are mostly based on cryptographic techniques, which often incur additional communication and computation overhead, while using Type I adversarial examples to hide users' privacy data is a brand-new exploration. Specifically, HRG utilizes the high-level representations of DNNs to guide generators, and GAG leverages the generative adversarial network to transform original images. Our solution does not involve any model modifications and allows DNNs to run directly on transformed data, thus arousing no additional communication and computation overhead. Extensive experiments on MNIST, CIFAR-10, and ImageNet show that HRG can perfectly hide images into noise and achieve similar accuracy as the original accuracy, and GAG can generate natural images that are completely different from the original images with a small loss of accuracy.
Trajectory similarity search (TSS) is a common operation for spatiotemporal data analysis. However, the existing TSS methods are mainly focused on GPS trajectories produced by moving objects such as vehicles. Further, these corresponding optimization strategies cannot be directly applied in the metro scenario because the metro passenger trajectory is totally different from the GPS trajectory. To fill this gap, we systematically analyze the unique spatiotemporal characteristics of metro passenger trajectories and propose a similarity search solution named PMMTss for the metro system. The core idea of this solution has two key points: first, we design a multi-layer index based on the spatiotemporal feature of metro trajectories, and all points of a trajectory are stored in this index. Second, we design a parallel multi-way merging-based trajectory similar search method, in which the candidate trajectory segments are merged and filtered. We evaluate this solution on a large dataset (Shenzhen Metro data for 3 consecutive months, 6.976 million trajectories with 260 million records). When lengths of input trajectories are 16, 32, and 64, respectively, the corresponding search times are 0.004 s, 0.016 s, and 0.036 s, respectively. Compared to the baseline PPJion+, the query times are reduced by 99.7%, 98.8%, and 97.6%, respectively.
Adversarial examples in which imperceptible perturbations to the input can easily subvert a well-trained model’s prediction pose huge potential security threats to deep neural networks (DNNs). As an effective way to resist adversarial samples, input reconstruction can eliminate the antagonism of adversarial examples in the inference process without involving modifications to the target model’s structure and parameters. However, preprocessing inputs often results in some loss of the protected model’s prediction accuracy. In this paper, we introduce a new input reconstruction method that adopts the high-level representation difference constraint and the relative reconstruction constraint on a dual autoencoder to advance the prediction accuracy of the protected model. The high-level representation difference constraint utilizes the gap between the protected model’s high-level representations, activated by clean images, and their adversarial examples to guide the training of the dual autoencoder. Additionally, the relative reconstruction constraint is imposed on latent representations and their noisy versions to advance the robustness of the dual autoencoder to tiny perturbations. The extensive empirical experiments on two real datasets, CIFAR-10 and ImageNet, show that the presented approach demonstrates exceptional performance in resisting different types of attacks.