Models based on deep learning are prone to misjudging the results when faced with adversarial examples. In this paper, we propose an MCTS-T algorithm for generating adversarial examples of cross-site scripting (XSS) attacks based on Monte Carlo tree search (MCTS) algorithm. The MCTS algorithm enables the generation model to provide a reward value that reflects the probability of generative examples bypassing the detector. To guarantee the antagonism and feasibility of the generative adversarial examples, the bypassing rules are restricted. The experimental results indicate that the missed detection rate of adversarial examples is significantly improved after the MCTS-T generation algorithm. Additionally, we construct a generative adversarial network (GAN) to optimize the detector and improve the detection rate when dealing with adversarial examples. After several epochs of adversarial training, the accuracy of detecting adversarial examples is significantly improved.
Aiming at unknown or variant ransomware attack encrypted with SSL (Secure Sockets Layer)/ TLS (Transport Layer Security) protocol, a detection framework named TGAN-IDS (Transferred Generating Adversarial Network-Intrusion Detection System) based on dual generative adversarial networks is presented in this paper. In this framework, DCGAN (Deep Convolutional Generative Adversarial Network) is adopted to train a generator which has good performance to generate adversarial sample, and is transferred to the generator of TGAN. A pre-training model named PreD is built based on CNN (Convolutional Neural Network), which has good performance to do binary classification, and is transferred to the discriminator of TGAN. The generator and discriminator of TGAN play games in training process until the discriminator has a strong ability to detection unknown attack, and then it is output as an anomaly detector. In order to suppress the deterioration of normal sample detection ability during adversarial training of TGAN, a reconstruction loss function is introduced into the target function of TGAN. Experiments on a mixed dataset which is constructed by CICIDS2017 and other ransomware datasets show comparing with other deep learning network, such as AlexNet, ResNet and DenseNet etc., TGAN-IDS performs well in the indicators of detection accuracy, recall or F1-score etc. Also experiments on KDD99, SWaT and WADI datasets show that TGAN-IDS is suitable for other unencrypted unknown network attack detection.
Detecting Android malware in its spread or download stage is a challenging work, which can realize early detection of malware before it reaches user side. In this paper, we propose a two-stage detection framework based on feature enhancement and cascade deep forest. This method can detect the traffic generated in the encrypted transmission process of Android malware. The first stage realizes the binary classification of benign and malicious software. The second stage realizes the multi-classification of different categories of malware. To enhance data representation, convolutional neural networks is used to extract benign and malicious features in the first stage, and the principal component analysis method is used to extract the malicious features in the second stage. Theses extracted features are spliced with the payload part of the traffic to form fusion features for classification task. In order to adapt to different scale of samples, especially for the small-scale sample, cascaded deep forest method is proposed to construct the classification model. In this model, many layers that consist of base classifiers are cascaded and the number of layers can be automatically adjusted according to the scale of the samples. With different combinations of base classifiers in each layer, the optima detection accuracy is archived in the two stages. The experimental results on several datasets prove that the proposed method is effective for encrypted transmission detection of Android malware. It is also suitable for the detection of unknown attacks.
1 May 2020 in International Journal of Advanced Robotic Systems
Path planning of lunar robots is the guarantee that lunar robots can complete tasks safely and accurately. Aiming at the shortest path and the least energy consumption, an adaptive potential field ant colony algorithm suitable for path planning of lunar robot is proposed to solve the problems of slow convergence speed and easy to fall into local optimum of ant colony algorithm. This algorithm combines the artificial potential field method with ant colony algorithm, introduces the inducement heuristic factor, and adjusts the state transition rule of the ant colony algorithm dynamically, so that the algorithm has higher global search ability and faster convergence speed. After getting the planned path, a dynamic obstacle avoidance strategy is designed according to the predictable and unpredictable obstacles. Especially a geometric method based on moving route is used to detect the unpredictable obstacles and realize the avoidance of dynamic obstacles. The experimental results show that the improved adaptive potential field ant colony algorithm has higher global search ability and faster convergence speed. The designed obstacle avoidance strategy can effectively judge whether there will be collision and take obstacle avoidance measures.
Spatial transcriptomics is an emerging technology that enables the profiling of gene expression in tissues while preserving spatial location information. This innovative approach is anticipated to provide a comprehensive understanding of the spatial distribution of different cells within tissues and facilitate in-depth analysis of tissue structure. To accurately recognize spatial domains from spatial transcriptomics, we have introduced a generalized deep learning method called GAAEST (Graph Attention-based Autoencoder for Spatial Transcriptomics). Our proposed approach effectively integrates both spatial location information and gene expression data from spatial transcriptomics. Specifically, it leverages spatial location details to construct a neighborhood graph and employs a graph attention network-based encoder to embed gene expression information into a spatially informed space. At the same time, to further optimize the learned potential embedding, self-supervised contrastive learning is introduced to capture spatial information at three levels: local, global and contextual feature of spots. Finally, the decoder reconstructs gene expressions, which are then clustered to identify spatial domains with similar expression patterns and spatial proximity. Based on our experiments conducted on multiple datasets, GAAEST consistently outperforms existing state-of-the-art methods. The proposed GAAEST demonstrates excellent capabilities in spatial domain recognition, positioning it as an ideal tool for advancing spatial transcriptomics research.