Up to now, most existing steganalytic methods are designed for grayscale images, and they are not suitable for color images that are widely used in current social networks. In this paper, we design a universal color i...
详细信息
In terms of the generative process, the Gamma-Gamma-Poisson Process (G2PP) is equivalent to the nonparametric topic model of Hierarchical Dirichlet Process (HDP). Considering the high computational cost of estimating ...
In terms of the generative process, the Gamma-Gamma-Poisson Process (G2PP) is equivalent to the nonparametric topic model of Hierarchical Dirichlet Process (HDP). Considering the high computational cost of estimating parameters in HDP, a parallel G2PP was developed to generate topics efficiently via multi-threading. Unfortunately, the above model needs to predefine the number of topics. To address this issue, we first propose a Topic Self-Adaptive Model (TSAM) for nonparametric and parallel topic discovery. In TSAM, a monitor-executor mechanism is developed to manage the global topic information using a hierarchical structure of threads. Based on the apparatus of copulas, we further extend our TSAM to TSAMcop for coherent topic modeling by exploiting a copula guided parallel Gibbs sampling algorithm. Extensive experiments validate the effectiveness of both TSAM and TSAMcop.
The convolutional neural network (CNN) is vulnerable to degraded images with even very small variations (e.g. corrupted and adversarial samples). One of the possible reasons is that CNN pays more attention to the most...
详细信息
The convolutional neural network (CNN) is vulnerable to degraded images with even very small variations (e.g. corrupted and adversarial samples). One of the possible reasons is that CNN pays more attention to the most discriminative regions, but ignores the auxiliary features when learning, leading to the lack of feature diversity for final judgment. In our method, we propose to dynamically suppress significant activation values of CNN by group-wise inhibition, but not fixedly or randomly handle them when training. The feature maps with different activation distribution are then processed separately to take the feature independence into account. CNN is finally guided to learn richer discriminative features hierarchically for robust classification according to the proposed regularization. Our method is comprehensively evaluated under multiple settings, including classification against corruptions, adversarial attacks and low data regime. Extensive experimental results show that the proposed method can achieve significant improvements in terms of both robustness and generalization performances, when compared with the state-of-the-art methods. Code is available at https://***/LinusWu/TENET_Training.
Collaborative Metric Learning (CML) has recently emerged as a popular method in recommendation systems (RS), closing the gap between metric learning and Collaborative Filtering. Following the convention of RS, existin...
详细信息
Group recommendation involves comprehensively considering various aspects, including members and items, to predict the overall interests of a group and recommend suitable items through a recommendation system. With th...
Group recommendation involves comprehensively considering various aspects, including members and items, to predict the overall interests of a group and recommend suitable items through a recommendation system. With the rapid evolution of the Internet, online group activities have become increasingly prevalent, making group recommendation a highly discussed topic within the realm of recommendation systems. However, current research in group recommendation still confronts the following challenges. Firstly, a predominant portion of group recommendation research focuses solely on aggregating group-level information, neglecting the valuable contribution of higher-order member-level insights to group recommendation. Moreover, some aggregation strategies overly prioritize fairness, thereby disregarding common real-world patterns. Secondly, previous studies have primarily relied on aggregating individual member interests to establish group preferences, lacking a holistic consideration of group dynamics. To tackle these challenges, this study introduces an innovative approach by implementing multi-channel hypergraph convolution within the member perspective. This approach aims to effectively extract higher-order insights from members by utilizing a member information enhancement module. By establishing group interconnections based on similarity and employing an adaptive fusion network to amalgamate multiple viewpoints, a final representation is derived. Experimental results demonstrate that our proposed model surpasses baseline models by 3% to 4% in terms of hit rate and accuracy, validating the efficacy of our approach.
The remarkable success of Face Recognition (FR) systems has raised significant privacy concerns, particularly the potential for unauthorized tracking of individuals on social media. To mitigate these risks, various ad...
The remarkable success of Face Recognition (FR) systems has raised significant privacy concerns, particularly the potential for unauthorized tracking of individuals on social media. To mitigate these risks, various adversarial attack methods have been proposed to protect individuals from unauthorized identification by FR systems. In this work, we propose a novel A dversarial F ace generation method via D iffusion M odification (AFDM), designed for imperceptible and transferable attacks against FR systems. The AFDM operates in the diffusion latent space of the face image, leveraging DDIM inversion and the DDIM process to achieve precise modifications. To ensure high visual fidelity, we propose a self-attention structure loss and a perceptual loss, effectively constraining the extent of diffusion-based alterations. Additionally, to enhance transferability, we introduce a face attribute loss that aligns the attribute embeddings of the protected face with those of the target face. Extensive experiments demonstrate that the proposed AFDM achieves state-of-the-art performance in attacking various FR models, with an average Protect Success Rate (PSR) of 93.32% on the CeleA-HQ and LADN datasets, while maintaining a more natural-looking protected face image with an FID score of 22.96. Through our systematic analysis, we justify the efficacy and generality of the proposed method.
GCC compiler is a retargetable compiler program that was developed to increase the efficiency of programs in the GNU system. In recent years, compiler optimization based on data dependency analysis has become an impor...
详细信息
To enhance robustness of complex networked systems, a simple method is introducing reinforced nodes which always function during failure propagation. A random scheme of node reinforcement can be considered as a benchm...
详细信息
Few-shot knowledge graph completion (FKGC) aims to query the unseen facts of a relation given its few-shot reference entity pairs. The side effect of noises due to the uncertainty of entities and triples may limit the...
详细信息
Humans excel at adapting perceptions and actions to diverse environments, enabling efficient interaction with the external world. This adaptive capability relies on the biological nervous system (BNS), which activates...
详细信息
暂无评论