Despite the recent successes of vanilla Graph Neural Networks (GNNs) on various tasks, their foundation on pairwise networks inherently limits their capacity to discern latent higher-order interactions in complex systems. To bridge this capability gap, we propose a novel approach exploiting the rich mathematical theory of simplicial complexes (SCs) - a robust tool for modeling higher-order interactions. Current SC-based GNNs are burdened by high complexity and rigidity, and quantifying higher-order interaction strengths remains challenging. Innovatively, we present a higher-order Flower-Petals (FP) model, incorporating FP Laplacians into SCs. Further, we introduce a Higher-order Graph Convolutional Network (HiGCN) grounded in FP Laplacians, capable of discerning intrinsic features across varying topological scales. By employing learnable graph filters, a parameter group within each FP Laplacian domain, we can identify diverse patterns where the filters' weights serve as a quantifiable measure of higher-order interaction strengths. The theoretical underpinnings of HiGCN's advanced expressiveness are rigorously demonstrated. Additionally, our empirical investigations reveal that the proposed model accomplishes state-of-the-art performance on a range of graph tasks and provides a scalable and flexible solution to explore higher-order interactions in graphs. Codes and datasets are available at https://github.com/Yiminghh/HiGCN.

Transformer models are widely used in AI applications such as Natural Language Processing (NLP), Computer Vision (CV), etc. However, enormous computation workload be-comes an obstacle to train large transformer models efficiently. Recently, some methods focus on reducing the computation workload during the training by skipping some layers. How-ever, these methods use simple probability distribution and coarse-grained probability calculation, which significantly affect the model accuracy. To address the issue, in this paper we propose a novel method to accelerate training—Sensitivity-Based Layer Dropping (SBLD). SBLD uses lay-er-wise sensitivity data to switch on/off transformer layers in proper order to keep high accuracy. Besides, we adjust the probability of skipping transformer layers with a scheduler to accelerate training speed and get faster convergence. Our results show that SBLD solves the accuracy drop issue com-pared with prior layer dropping methods. Our SBLD method can decrease end-to-end training time by 19.67% during training of GPT-3 Medium model, the same time increasing the accuracy by 1.65% w.r.t. baseline. Furthermore, for SwinV2-L model the obtained Top-1 and Top-5 accuracies are also higher vs. the baseline. Thus, the proposed method is efficient and practical to improve the large transformer model training.

Content Centric Network (CCN) can be extended to efficiently and reliably support content delivery and solve the network performance degradation caused by dynamic topology and intermittent connectivity of Vehicle Ad hoc NETwork (VANET). However, the in-network caching mechanism of Vehicular Content Centric Network (VCCN) is vulnerable against Cache Pollution Attack (CPA), where attackers aim to fill the buffer space with non-popular contents by releasing fake requests. Unavoidably, the cache hit ratio of content requests from legal users is degraded and the content retrieval latency is increased under CPA. Hence, it is critical to detect and mitigate CPA. The current solutions for static CCN cannot be directly applied into dynamic VCCN. In this article, we propose a detection scheme based on hybrid heterogeneous multi-classifier ensemble learning, where CPA is determined by the cooperation of multiple vehicles. In our scheme, each vehicle can build or join a cluster whose head possesses more common moving attributes of position, speed and direction with other members. Besides, the cluster head as a base learner is responsible for training its own classifier by making some relevant statistics on requests and hit ratio. Specifically, the problem of ensemble classifier making from the individual classifiers is formulated as a linear optimization problem, with the goal of minimizing the false ratio of detecting CPA. The generalization ability of ensemble learning can make very accurate predictions on CPA. By comparison, our detection scheme outperforms the existing schemes in terms of detection ratio, hit ratio, retrieval delay. Besides, simulations have proved that the overfitting problem of adopting a singe base learning algorithm can be alleviated in our scheme.

Named Data Networking (NDN) is one of the most promising information-centric networking architectures that can improve the network performance by supporting the large scale content distribution. However, the use of in-network caching mechanism increases the opportunity of cache pollution attack, where the attackers intend to reduce the cache hit of legal users by releasing fake requests to fill the precious cache with non-popular contents. To prevent the degradation of network performance caused by such an attack, it is becoming particularly important to detect the attack and then throttle it. In this article, we propose a detection and defense scheme with the help of grey forecast, which can effectively exploit the regularity of past Interests and popularity by comprehensively considering three major factors to predict the future popularity of each cached content. If the predicted popularity of any content differs too much from the actually calculated one in several consecutive slices, the pollution attack will be determined. Once the attack is detected, the defense will be taken by suppressing the popularity increase of the suspicious content to mitigate the damage of the pollution attack. We also consider a special case, where there exists a sudden burst of traffic from legal users that cannot be simply dropped. The simulations in ndnSIM indicate that our proposed method is effective in detecting and defending the pollution attack with higher cache hit, higher detecting ratio, and lower hop count compared to other state-of-the-art schemes.

Named data networking(NDN) is a very promising architecture for future network, which can improve the network performance due to its in-network caching feature. However, the pervasive caching is vulnerable against False-Locality Attack (FLA), one kind of cache pollution attack, where attackers repeatedly request a specific set of non-popular contents to replace popular contents. Therefore, the cache hit of legal requests is reduced and the response delay is increased. To mitigate this attack and improve the network performance, we propose a detection scheme based on Convolutional Neural Network (CNN) by fully exploiting the regularity of past requests. The input data of CNN are related to the inherent characteristics of the cached contents including the request ratio, the standard deviation of repeated Interests, the variance of request interval and the change of cache hit ratio. The output of CNN indicates whether FLA has been launched. Simulations through multi-topologies are conducted to validate the performance of our scheme. Compared with other state-of-the-art schemes, it is more effective in detecting FLA with higher detecting ratio, higher cache hit and lower hop count.