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.
Spatial co-location pattern is a subset of spatial features which shows association relationships based on the spatial neighborhoods. Because the previous prevalence measurements of a co-location pattern have not considered the visited information of spatial instances, co-location patterns do not reflect the social connections (such as their spatial instances are constantly visited by common or similar moving objects) between spatial features. In this paper, a special type of co-location pattern, “Highly visited co-location patterns”, is proposed, which considers the spatial proximity and visitor similarity of spatial features at the same time. A new measurement, “Minimum visitor similarity”, has been proposed to reflect the visitor similarity of co-location patterns. By discussing the properties of the minimum visitor similarity, we propose an efficient algorithm to mine the highly visited co-locations and give two pruning strategies to improve the efficiency of the algorithm. Finally, extensive experiments on YELP and Foursquare datasets prove the practicability and efficiency of the proposed algorithm, and we define a “Social Entropy” to prove that spatial features in the co-locations we mined have stronger social connections.