Applied Intelligence

Tang, Kaiyu; Fan, Yali; Song, Yan

doi: 10.1007/s10489-024-05529-4pmid: N/A

The problem of nonconvex robust tensor principal component analysis consists of recovering the low-rank and sparse part from a tensor corrupted by noise, which attracts a great deal of attention in a wide range of practical situations. However, existing nonconvex methods face a number of problems, the two most important of which are the restrictions on specific nonconvex functions and the information loss in low-rank part. In this paper, we propose a generalized nonconvex robust tensor principal component analysis model that includes some of the most popular nonconvex functions. Furthermore, we propose a partial weighted tensor nuclear norm and the corresponding partial weighted singular value thresholding operator to improve the generalized model, further reducing the loss of underlying tensor information while recovering the low rank tensor. Besides, two ADMM-based nonconvex algorithms are proposed to the generalized nonconvex model and the improved model respectively. We also analyze the convergence of the algorithms, the computational complexity, and the theoretical guarantee of the proposed models. Numerical experimental results on image data and video data show that our proposed models has superior performance compared to several state-of-the-art methods.

Sun, Lin; Liang, Hanbo; Ding, Weiping; Xu, Jiucheng; Chang, Baofang

doi: 10.1007/s10489-024-05555-2pmid: N/A

Swarm intelligence algorithms availably settle the issue of feature selection for classification, whereas the Mayfly Optimization Algorithm (MOA) proposed in recent years has the superiority of high precision, concise structure, and effortless enforcement; but it is still caught in local optimum, and the convergence speed needs to be optimized. Then, this work studies a new MOA and further develops a binary MOA to solve feature selection problems. Initially, based on two mapping mechanisms of Logistic-Tent and Cubic chaotic, the male and female populations of MOA are mapped respectively to enhance the variety of MOA populations in the exploration stage, and the Cubic chaotic mapping scheme is cited to dynamically disturb the global optimum to eliminate the limitation of easily getting into local optimum and promote local exploration ability for MOA. Secondly, the parameter fuzzy entropy is proposed to improve MOA, and then an adaptive function based on the parameter fuzzy entropy is constructed by using the historical optimal result of mayfly in the population of MOA. The parameter fuzzy entropy is used as an impact factor to variously adapt the inertia weight, balance global optimization and local exploration capability of population, and increase the variety of population and uniformity of distribution. Further, the contraction factor is improved to be introduced into MOA, and two learning factors are restricted by parameters, so that the velocity of mayfly individuals is not too large, and the convergence performance of MOA is effectively improved. Finally, the binary MOA is constructed based on the S-type transfer function, so that it can process those continuous data, and the optimal feature subset is selected by employing the fitness function. Simulation experimental results on 16 benchmark functions and 12 public datasets show that the binary MOA has great optimization performance, and the effectiveness of the designed feature selection algorithm has been verified.

Yang, Tian; Wang, Gang; Lai, Jian; Wang, Yang

doi: 10.1007/s10489-024-05527-6pmid: N/A

In many domains, trajectory prediction a crucial task. Uncertain information, such as complementary and correlated information between multiple features, complex interactive information, weather and temperature, increases the difficulty of trajectory prediction. In real life scenarios, multi-feature information often interact and complement each other. It is difficult to isolate multi-feature information and consider their effects separately. At present, many trajectory prediction methods based on multi-feature information fusion often ignore the correlation impact information between multiple features. This may lead to insufficient utilization of multi-feature information, which limits the improvement of prediction accuracy. In this paper, a new information fusion method based on uncertain information processing is proposed. To begin with, the local support impact between multi-feature information is modeled by the local support function of power average (P-A) operator. The global support impact between them is modeled by the global support function of P-A. Then, through the local support impact and global support impact, the weight of each feature information is obtained respectively. The nonlinear fusion is performed to make full use of multi-feature information. What is more, in the discriminator, a new fuzzy soft discrimination state is added. The fuzzy soft discrimination state improves the flexibility of the discriminator. At last, based on the soft discrimination, a new loss function is proposed by combining the belief entropy and cross entropy. With the uncertain information processing, compared to some baselines, the proposed method achieves higher accuracy in some parts of the ETH/UCY dataset.

Lu, Yuting; Wang, Gaocai; Huang, Xianfei; Huang, Shuqiang; Wu, Man

doi: 10.1007/s10489-024-05540-9pmid: N/A

In the dynamic smart grid landscape, accurate probabilistic forecasting of electric load is critical. This paper presents a novel 24-hour-ahead probabilistic load forecasting model by integrating quantile regression with a parallel convolutional neural network (CNN) and bidirectional gated recurrent unit (BiGRU) architecture. Carefully tuning hyperparameters can enhance model performance and generalization capability. Consequently, we propose an improved whale optimization algorithm for automatic hyperparameter tuning of the forecasting model. Case studies demonstrate the proposed method’s superior performance over benchmark models in terms of average interval score and pinball loss. In addition, it exhibits valid coverage and tight interval bandwidths. The model provides precise short-term load forecasts to support robust smart grid planning and operations.

Shadrooh, Shiva; Nørvåg, Kjetil

doi: 10.1007/s10489-024-05545-4pmid: N/A

Smurfing in financial networks is a popular fraud technique in which fraudsters inject their illegal money into the legitimate financial system. This activity is performed within a short period of time, with recurring transactions and multiple intermediaries. A major problem of existing graph-based methods for detecting smurfing is that they fall short of retrieving accurate fraud patterns. Consequently, the result is numerous non-fraudulent patterns alongside a few fraud patterns, causing a high false-positive rate. To alleviate this problem, we propose SMoTeF, a framework that extends existing graph-based smurf detection methods by distinguishing fraudulent smurfing patterns from non-fraudulent ones, thus significantly reducing the false-positive ratio. The core of the approach is a novel algorithm based on computing maximum temporal flow within temporal order of events. In order to evaluate the approach, a framework for injecting various smurfing patterns is developed, and experimental results on three real-world datasets from different domains show that SMoTeF significantly improves on the effectiveness of the state-of-the-art baseline, with only marginal runtime overhead.

Fang, Jie; Wu, Zhichao; Xu, Mengyun; Chen, Hongting

doi: 10.1007/s10489-024-05503-0pmid: N/A

Traffic forecasting using deep learning represents a crucial aspect of intelligent transportation systems, carrying substantial implications for congestion reduction and efficient route planning. Despite its significance, accurately predicting traffic states remains a challenge. Existing methodologies focus on capturing the temporal trends of traffic states and the spatial dependencies between roads to enhance prediction accuracy. However, two noteworthy limitations persist in these approaches: (1) Many models neglect the interaction between spatiotemporal features across varying time spans, hindering their ability to utilize traffic state information effectively for predicting future conditions. (2) Genuine correlations between roads are time-varying, making it inadequate to rely on static graphs or static pre-trained node embeddings to model dynamic correlations between roads. To address these challenges, we propose the Multiple Time-Scale Graph Attention Network (MTS-GATN), which comprises two key modules: the Multiple Time-Scale Spatiotemporal Features Extraction Module and the Feature Augmentation Module. The first module involves stacking multiple spatiotemporal extraction layers to discern traffic state information at different time scales. In the second module, we employ dynamic spatial semantic embedding for feature augmentation, providing nodes with dynamic representations over time. Subsequently, we leverage a multi-head spatiotemporal attention mechanism to comprehensively consider location information and real-time semantic data, facilitating the interaction of traffic state information across multiple time scales. Experimental results on two distinct traffic datasets validate the superior performance of MTS-GATN in medium-term and long-term forecasting scenarios.

Kong, Weiheng; Sun, Minghui

doi: 10.1007/s10489-024-05577-wpmid: N/A

In this paper, we proposed a novel deep image coding and decoding model for GMMQI (Gaussian mixture model quartile intervals, GMMQI) and a variable rate bit allocation method to optimize the rate distortion performance and prioritize the transmissions of more significant information. First, to address the problem of bit streams leading to ambiguity during encoding and decoding, we convert the image into a potential tensor, each element of which uses a four-bit parameter dictionary to preserve the parameter bits. Then, the variable rate is calculated using quaternions based on the parameter dictionary, and a hybrid CLM &BAM (Channel latent map & Bit allocation map) approach is designed to assign bits to the potential tensor and encode it, which transforms the problem of finding the optimal encoder-decoder into finding the optimal hyper-parameters in the model and reduces the complexity of the GMMQI model. Finally, a GMMQI approach with variable rate bit allocation is developed in combination with CLM &BAM, to be able to prioritize the transmission of more significant information. The experimental results show that the GMMQI method reaches an advanced level compared to the traditional image compression standards BPG, JPEG2000, and JPEG, and is comparable to the most advanced level compared to the existing deep learning based compression methods.

Fang, Wanchuan; Wang, Chuansheng; Li, Zuoyong; Grau, Antoni; Lai, Taotao; Chen, Jianzhang

doi: 10.1007/s10489-024-05569-wpmid: N/A

Recently, CNN-based all-in-one image enhancement methods have been proposed to solve multiple image degradation tasks. However, these CNN-based methods usually have two limitations. One limitation is that they usually design a specific encoder for each image enhancement task, lacking of a unified and simple framework. The other limitation is that they can not effectively capture global image degradation information, as use of the CNN-based encoders has a limited local receptive field. In this work, we propose a TransEncoder-based All-in-one Image Enhancement Network (TAENet), with a single encoder and a decoder, for simultaneously handling multiple image enhancement tasks. Specifically, we propose a Transformer-based Encoder (TransEncoder), which introduces instance normalization to transformer for color recovery. The TransEncoder model global degradation information by using the transformer’s global self-attention mechanism. Additionally, inspired by the Mie scattering model, we propose a novel depth loss function for perceiving image depth information by minimizing the depth difference between the enhanced image and the ground-truth, thus further improving model performance. Moreover, a novel contrastive loss is introduced to strengthen task-generalization performance by enhancing the model’s representation capability. Experiments show that the TAENet outperforms 24 state-of-the-art methods on image dehazing, image deraining, and low-light image enhancement.

Tang, Zhe; Tang, Jing; Zou, Dengpeng; Rao, Junyi; Qi, Fang

doi: 10.1007/s10489-024-05559-ypmid: N/A

Camouflaged object detection (COD) entails identifying objects in an image that blend with the background. However, most traditional COD methods have not comprehensively considered the information provided by the overall region and edges of the objects. To address this problem, a new two-guidance joint network based on coarse and edge maps is proposed for COD. Particularly, an information guidance module is designed to inject edges and overall information into the network’s backbone features. Meanwhile, a feature observation model based on skip connections and multi-scale perception is designed to capture multi-scale image details and structures. To avoid the loss of semantic information in low-level features, a full-image attention mechanism is designed to integrate high-level features into low-level features, thereby improving the resolution of the object masks. We compared the proposed network with state-of-the-art models on three well-known datasets, and the experimental results show the proposed network has significant improvement. By exploring valuable boundary information and overall object information, the proposed network can segment object edges while also considering the segmentation effect of the entire object. Our code has been open-sourced at https://github.com/Huah2019/TJNet.