Mingsheng Shang

Liangming Chen, Long Jin, Mingsheng Shang

The question of what kind of convolutional neural network (CNN) structure performs well is fascinating. In this work, we move toward the answer with one more step by connecting zero stability and model performance. Specifically, we found that if a discrete solver of an ordinary differential equation is zero stable, the CNN corresponding to that solver performs well. We first give the interpretation of zero stability in the context of deep learning and then investigate the performance of existing first- and second-order CNNs under different zero-stable circumstances. Based on the preliminary observation, we provide a higher-order discretization to construct CNNs and then propose a zero-stable network (ZeroSNet). To guarantee zero stability of the ZeroSNet, we first deduce a structure that meets consistency conditions and then give a zero stable region of a training-free parameter. By analyzing the roots of a characteristic equation, we theoretically obtain the optimal coefficients of feature maps. Empirically, we present our results from three aspects: We provide extensive empirical evidence of different depth on different datasets to show that the moduli of the characteristic equation's roots are the keys for the performance of CNNs that require historical features; Our experiments show that ZeroSNet outperforms existing CNNs which is based on high-order discretization; ZeroSNets show better robustness against noises on the input. The source code is available at \url{https://github.com/LongJin-lab/ZeroSNet}.

Hongbin Liang, Hezhe Qiao, Wei Huang et al.

Ensuring the safety of vulnerable road users through accurate prediction of pedestrian crossing intention (PCI) plays a crucial role in the context of autonomous and assisted driving. Analyzing the set of observation video frames in ego-view has been widely used in most PCI prediction methods to forecast the cross intent. However, they struggle to capture the critical events related to pedestrian behaviour along the temporal dimension due to the high redundancy of the video frames, which results in the sub-optimal performance of PCI prediction. Our research addresses the challenge by introducing a novel approach called \underline{T}emporal-\underline{c}ontextual Event \underline{L}earning (TCL). The TCL is composed of the Temporal Merging Module (TMM), which aims to manage the redundancy by clustering the observed video frames into multiple key temporal events. Then, the Contextual Attention Block (CAB) is employed to adaptively aggregate multiple event features along with visual and non-visual data. By synthesizing the temporal feature extraction and contextual attention on the key information across the critical events, TCL can learn expressive representation for the PCI prediction. Extensive experiments are carried out on three widely adopted datasets, including PIE, JAAD-beh, and JAAD-all. The results show that TCL substantially surpasses the state-of-the-art methods. Our code can be accessed at https://github.com/dadaguailhb/TCL.

Mei Liu, Liangming Chen, Xiaohao Du et al.

Deep neural networks often suffer from poor performance or even training failure due to the ill-conditioned problem, the vanishing/exploding gradient problem, and the saddle point problem. In this paper, a novel method by acting the gradient activation function (GAF) on the gradient is proposed to handle these challenges. Intuitively, the GAF enlarges the tiny gradients and restricts the large gradient. Theoretically, this paper gives conditions that the GAF needs to meet, and on this basis, proves that the GAF alleviates the problems mentioned above. In addition, this paper proves that the convergence rate of SGD with the GAF is faster than that without the GAF under some assumptions. Furthermore, experiments on CIFAR, ImageNet, and PASCAL visual object classes confirm the GAF's effectiveness. The experimental results also demonstrate that the proposed method is able to be adopted in various deep neural networks to improve their performance. The source code is publicly available at https://github.com/LongJin-lab/Activated-Gradients-for-Deep-Neural-Networks.

Tingyu Song, Guo Gan, Mingsheng Shang et al.

We introduce IFIR, the first comprehensive benchmark designed to evaluate instruction-following information retrieval (IR) in expert domains. IFIR includes 2,426 high-quality examples and covers eight subsets across four specialized domains: finance, law, healthcare, and science literature. Each subset addresses one or more domain-specific retrieval tasks, replicating real-world scenarios where customized instructions are critical. IFIR enables a detailed analysis of instruction-following retrieval capabilities by incorporating instructions at different levels of complexity. We also propose a novel LLM-based evaluation method to provide a more precise and reliable assessment of model performance in following instructions. Through extensive experiments on 15 frontier retrieval models, including those based on LLMs, our results reveal that current models face significant challenges in effectively following complex, domain-specific instructions. We further provide in-depth analyses to highlight these limitations, offering valuable insights to guide future advancements in retriever development.

Khushnood Abbas, Alireza Abbasi, Dong Shi et al.

Understanding the evolutionary patterns of real-world evolving complex systems such as human interactions, transport networks, biological interactions, and computer networks has important implications in our daily lives. Predicting future links among the nodes in such networks reveals an important aspect of the evolution of temporal networks. To analyse networks, they are mapped to adjacency matrices, however, a single adjacency matrix cannot represent complex relationships (e.g. temporal pattern), and therefore, some approaches consider a simplified representation of temporal networks but in high-dimensional and generally sparse matrices. As a result, adjacency matrices cannot be directly used by machine learning models for making network or node level predictions. To overcome this problem, automated frameworks are proposed for learning low-dimensional vectors for nodes or edges, as state-of-the-art techniques in predicting temporal patterns in networks such as link prediction. However, these models fail to consider temporal dimensions of the networks. This gap motivated us to propose in this research a new node embedding technique which exploits the evolving nature of the networks considering a simple three-layer graph neural network at each time step, and extracting node orientation by Given's angle method. To prove our proposed algorithm's efficiency, we evaluated the efficiency of our proposed algorithm against six state-of-the-art benchmark network embedding models, on four real temporal networks data, and the results show our model outperforms other methods in predicting future links in temporal networks.

Khushnood Abbas

Novelty attracts attention like popularity. Hence predicting novelty is as important as popularity. Novelty is the side effect of competition and aging in evolving systems. Recent behavior or recent link gain in networks plays an important role in emergence or trend. We exploited this wisdom and came up with two models considering different scenarios and systems. Where recent behavior dominates over total behavior (total link gain) in the first one, and recent behavior is as important as total behavior for future link gain in second one. It suppose that random walker walks on a network and can jump to any node, the probablity of jumping or making connection to other node is based on which node is recently more active or receiving more links. In our assumption random walker can also jump to node which is already popular but recently not popular. We are able to predict rising novelties or popular nodes which is generally suppressed under preferential attachment effect. To show performance of our model we have conducted experiments on four real data sets namely, MovieLens, Netflix, Facebook and Arxiv High Energy Physics paper citation. For testing our model we used four information retrieval indices namely Precision, Novelty, Area Under Receiving Operating Characteristic(AUC) and Kendal's rank correlation coefficient. We have used four benchmark models for validating our proposed models. Although our model doesn't perform better in all the cases but, it has theoretical significance in working better for recent behavior dominant systems.

Khushnood Abbas, Mingsheng Shang, Cai Shi-Min et al.

Identifying emerging influential or popular node/item in future on network is a current interest of the researchers. Most of previous works focus on identifying leaders in time evolving networks on the basis of network structure or node's activity separate way. In this paper, we have proposed a hybrid model which considers both, node's structural centrality and recent activity of nodes together. We consider that the node is active when it is receiving more links in a given recent time window, rather than in the whole past life of the node. Furthermore our model is flexible to implement structural rank such as PageRank and webpage click information as activity of the node. For testing the performance of our model, we adopt the PageRank algorithm and linear preferential attachment based model as the baseline methods. Experiments on three real data sets (i.e Movielens, Netflix and Facebook wall post data set), we found that our model shows better performance in terms of finding the emerging influential nodes that were not popular in past.

Yao-Dong Zhao, Shi-Min Cai, Ming Tang et al.

The tripartite graph is one of the commonest topological structures in social tagging systems such as Delicious, which has three types of nodes (i.e., users, URLs and tags). Traditional recommender systems developed based on collaborative filtering for the social tagging systems bring very high demands on CPU time cost. In this paper, to overcome this drawback, we propose a novel approach that extracts non-overlapping user clusters and corresponding overlapping item clusters simultaneously through coarse clustering to accelerate the user-based collaborative filtering and develop a fast recommendation algorithm for the social tagging systems. The experimental results show that the proposed approach is able to dramatically reduce the processing time cost greater than $90\%$ and relatively enhance the accuracy in comparison with the ordinary user-based collaborative filtering algorithm.

Wei Zeng, An Zeng, Hao Liu et al.

With the rapid growth of the Internet and overwhelming amount of information that people are confronted with, recommender systems have been developed to effiectively support users' decision-making process in online systems. So far, much attention has been paid to designing new recommendation algorithms and improving existent ones. However, few works considered the different contributions from different users to the performance of a recommender system. Such studies can help us improve the recommendation efficiency by excluding irrelevant users. In this paper, we argue that in each online system there exists a group of core users who carry most of the information for recommendation. With them, the recommender systems can already generate satisfactory recommendation. Our core user extraction method enables the recommender systems to achieve 90% of the accuracy by taking only 20% of the data into account.

Wei Zeng, An Zeng, Ming-Sheng Shang et al.

With the rapid growth of the Internet and overwhelming amount of information and choices that people are confronted with, recommender systems have been developed to effectively support users' decision-making process in the online systems. However, many recommendation algorithms suffer from the data sparsity problem, i.e. the user-object bipartite networks are so sparse that algorithms cannot accurately recommend objects for users. This data sparsity problem makes many well-known recommendation algorithms perform poorly. To solve the problem, we propose a recommendation algorithm based on the semi-local diffusion process on a user-object bipartite network. The numerical simulation on two sparse datasets, Amazon and Bookcross, show that our method significantly outperforms the state-of-the-art methods especially for those small-degree users. Two personalized semi-local diffusion methods are proposed which further improve the recommendation accuracy. Finally, our work indicates that sparse online systems are essentially different from the dense online systems, all the algorithms and conclusions based on dense data should be rechecked again in sparse data.

Wei Zeng, An Zeng, Ming-Sheng Shang et al.

During the past a few years, users' membership in the online system (i.e. the social groups that online users joined) are wildly investigated. Most of these works focus on the detection, formulation and growth of online communities. In this paper, we study users' membership in a coupled system which contains user-group and user-object bipartite networks. By linking users' membership information and their object selection, we find that the users who have collected only a few objects are more likely to be "influenced" by the membership when choosing objects. Moreover, we observe that some users may join many online communities though they collected few objects. Based on these findings, we design a social diffusion recommendation algorithm which can effectively solve the user cold-start problem. Finally, we propose a personalized combination of our method and the hybrid method in [PNAS 107, 4511 (2010)], which leads to a further improvement in the overall recommendation performance.

Qian-Ming Zhang, An Zeng, Ming-Sheng Shang

Information overload is a serious problem in modern society and many solutions such as recommender system have been proposed to filter out irrelevant information. In the literature, researchers mainly dedicated to improve the recommendation performance (accuracy and diversity) of the algorithms while overlooked the influence of topology of the online user-object bipartite networks. In this paper, we find that some information provided by the bipartite networks is not only redundant but also misleading. With such "less can be more" feature, we design some algorithms to improve the recommendation performance by eliminating some links from the original networks. Moreover, we propose a hybrid method combining the time-aware and topology-aware link removal algorithms to extract the backbone which contains the essential information for the recommender systems. From the practical point of view, our method can improve the performance and reduce the computational time of the recommendation system, thus improve both of their effectiveness and efficiency.