Search

Abstract Navigating mobile robots in crowded environments poses a significant challenge and is essential for the coexistence of robots and humans in future intelligent societies. As a pragmatic data-driven approach, deep reinforcement learning (DRL) holds promise for addressing this challenge. However, current DRL-based navigation methods have possible improvements in understanding agent interactions, feedback mechanism design, and decision foresight in dynamic environments. This paper introduces the model inductive bias enhanced deep reinforcement learning (MIBE-DRL) method, drawing inspiration from a fusion of data-driven and model-driven techniques. MIBE-DRL extensively incorporates model inductive bias into the deep reinforcement learning framework, enhancing the efficiency and safety of robot navigation. The proposed approach entails a multi-interaction network featuring three modules designed to comprehensively understand potential agent interactions in dynamic environments. The pedestrian interaction module can model interactions among humans, while the temporal and spatial interaction modules consider agent interactions in both temporal and spatial dimensions. Additionally, the paper constructs a reward system that fully accounts for the robot’s direction and position factors. This system's directional and positional reward functions are built based on artificial potential fields (APF) and navigation rules, respectively, which can provide reasoned evaluations for the robot's motion direction and position during training, enabling it to receive comprehensive feedback. Furthermore, the incorporation of Monte-Carlo tree search (MCTS) facilitates the development of a foresighted action strategy, enabling robots to execute actions with long-term planning considerations. Experimental results demonstrate that integrating model inductive bias significantly enhances the navigation performance of MIBE-DRL. Compared to state-of-the-art methods, MIBE-DRL achieves the highest success rate in crowded environments and demonstrates advantages in navigation time and maintaining a safe social distance from humans.

The anti-detection capabilities of adversarial malware examples have drawn the attention of antivirus vendors and researchers. In black-box scenarios where internal information of the target model cannot be accessed, with the ability of the reinforcement learning (RL) agents to adjust strategies based on the feedback from the environment, existing RL-based methods enable evasion of Windows PE malware detectors. However, obtaining evasion rewards as positive feedback in the black-box setting proves challenging, resulting in low training efficiency. To address this issue, we introduce the intrinsic curiosity reward into the framework to motivate the agent to explore unknown state spaces and learn effective evasion strategies. Additionally, we employ the generative adversarial network (GAN) to obtain varying synthetic data, which replaces random or benign bytes as adversarial payloads for the agent's action content, improving attack capabilities and reducing the risk of hard-coded adversarial perturbations being anchored. We compare the attack performance with other RL-based baseline methods, and experimental results show that our framework is more flexible and effective, achieving a 63%-85% attack success rate against EMBER, FireEye and MalConv. Even if defensive measures are taken, the proposed method still has a certain attack capability, and the success rate remains between 48%-67%.

The object detection of synthetic aperture radar (SAR) ships holds significant promise for water traffic monitoring, ship search and rescue, and maritime warning tasks. Regrettably, most of the existing SAR ship object detection is limited to the fully supervised paradigm, exhibiting a firm reliance on data labels, and inherent challenges such as multiscale ship feature disparities and indistinct small-sized ships also make SAR ship detection difficult. To address these, we propose a scale-mixing enhanced and dual consistency guided semisupervised object detection (SMDC-SSOD) method. Specifically, this method is based on the teacher–student framework and primarily comprises three core components: cross-scale feature mixing (CSFM) scheme, scale change consistency guidance (SCCG) strategy, and proposal consistency guidance (PCG) strategy, which can efficiently conduct end-to-end semisupervised learning from limited data labeling, achieving low-cost and high-performance ship perception. CSFM scheme includes interpyramid and intrapyramid feature cross-scale mixings, which can improve the network's adaptability for multiscale ship characteristics and increase focus on small-sized ships. SCCG strategy leverages variations in confidence scores at different scales to select valuable pseudolabels, providing more precise guidance for the student network. PCG strategy further reflects the positioning quality of pseudolabels through the proposal consistency generated by the student network, guiding it to make high-quality predictions. The experimental results on the publicly available HRSID, BBox-SSDD, and SAR-Ship-Dataset demonstrate that SMDC-SSOD can accurately detect SAR ships with an extremely low data annotation rate (below 10%) and achieve optimal detection performance compared to state-of-the-art methods.

Abstract It is a key point to find out the actual privileges of network users in network security risk assessment. The Privilege dependency graph (PDG) provides an effective way to reason the actual privileges of network users from their initial privileges. The existing User Actual Privilege reasoning method is time‐consuming and not suitable for large‐scale networks. This paper introduces a fast User Actual Privilege reasoning framework based on PDG reduction. The core idea is to reduce the size of the graph as much as possible before the actual privilege reasoning. Three different nodes merged scenarios are proposed and discussed, as well as the influences of different execution sequences and execution times. Networks of different sizes were simulated to validate the effectiveness and scalability of their method. The experimental results show that the proposed method can decrease the time of User Actual Privilege reasoning by over 25% in large‐scale networks.

Deep neural networks (DNNs) are increasingly being used in malware detection and their robustness has been widely discussed. Conventionally, the development of an adversarial example generation scheme for DNNs involves either detailed knowledge concerning the model (i.e., gradient-based methods) or a substantial quantity of data for training a surrogate model. However, under many real-world circumstances, neither of these resources is necessarily available. Our work introduces the concept of the instance-based attack, which is both interpretable and suitable for deployment in a black-box environment. In our approach, a specific binary instance and a malware classifier are utilized as input. By incorporating data augmentation strategies, sufficient data are generated to train a relatively simple and interpretable model. Our methodology involves providing explanations for the detection model, which entails displaying the weights assigned to different components of the specific binary. Through the analysis of these explanations, we discover that the data subsections have a significant impact on the identification of malware. In this study, a novel function preserving transformation algorithm designed specifically for data subsections is introduced. Our approach involves leveraging binary diversification techniques to neutralize the effects of the most heavily-weighted section, thus generating effective adversarial examples. Our algorithm can fool the DNNs in certain cases with a success rate of almost 100%. Instance attack exhibits superior performance compared to the state-of-the-art approach. Notably, our technique can be implemented in a black-box environment and the results can be verified utilizing domain knowledge. The model can help to improve the robustness of malware detectors.

Instance segmentation in remote sensing (RS) imagery aims to predict the locations of instances and represent them with pixel-level masks. Thanks to the more accurate pixel-level information for each instance, instance segmentation has enormous potential applications in resource planning, urban surveillance, and military reconnaissance. However, current RS imagery instance segmentation methods mostly follow the fully supervised paradigm, relying on expensive pixel-level labels. Moreover, remote sensing imagery suffers from cluttered backgrounds and significant variations in target scales, making segmentation challenging. To accommodate these limitations, we propose a semantic attention enhancement and structured model-guided multi-scale weakly supervised instance segmentation network (SASM-Net). Building upon the modeling of spatial relationships for weakly supervised instance segmentation, we further design the multi-scale feature extraction module (MSFE module), semantic attention enhancement module (SAE module), and structured model guidance module (SMG module) for SASM-Net to enable a balance between label production costs and visual processing. The MSFE module adopts a hierarchical approach similar to the residual structure to establish equivalent feature scales and to adapt to the significant scale variations of instances in RS imagery. The SAE module is a dual-stream structure with semantic information prediction and attention enhancement streams. It can enhance the network’s activation of instances in the images and reduce cluttered backgrounds’ interference. The SMG module can assist the SAE module in the training process to construct supervision with edge information, which can implicitly lead the model to a representation with structured inductive bias, reducing the impact of the low sensitivity of the model to edge information caused by the lack of fine-grained pixel-level labeling. Experimental results indicate that the proposed SASM-Net is adaptable to optical and synthetic aperture radar (SAR) RS imagery instance segmentation tasks. It accurately predicts instance masks without relying on pixel-level labels, surpassing the segmentation accuracy of all weakly supervised methods. It also shows competitiveness when compared to hybrid and fully supervised paradigms. This research provides a low-cost, high-quality solution for the instance segmentation task in optical and SAR RS imagery.

With the growing scale of pre-trained language models (PLMs), full parameter fine-tuning becomes prohibitively expensive and practically infeasible. Therefore, parameter-efficient adaptation techniques for PLMs have been proposed to learn through incremental updates of pre-trained weights, such as in low-rank adaptation (LoRA). However, LoRA relies on heuristics to select the modules and layers to which it is applied, and assigns them the same rank. As a consequence, any fine-tuning that ignores the structural information between modules and layers is suboptimal. In this work, we propose structure-aware low-rank adaptation (SaLoRA), which adaptively learns the intrinsic rank of each incremental matrix by removing rank-0 components during training. We conduct comprehensive experiments using pre-trained models of different scales in both task-oriented (GLUE) and task-agnostic (Yelp and GYAFC) settings. The experimental results show that SaLoRA effectively captures the structure-aware intrinsic rank. Moreover, our method consistently outperforms LoRA without significantly compromising training efficiency.

Abstract Deep neural networks are vulnerable to adversarial examples. However, existing attacks exhibit relatively low efficacy in generating transferable adversarial examples. Improved transferability to address this issue is proposed via a rotation‐invariant attack method that maximizes the loss function w.r.t the random rotated image instead of the original input at each iteration, thus mitigating the high correlation between the adversarial examples and the source models and making the adversarial examples more transferable. Extensive experiments show that the proposed method can significantly improve the transferability of the adversarial examples with almost no extra computational cost and can be integrated into various methods. In addition, when this method is easily applied through a plug‐in, the average attack success rate against six robustly trained models increases by 5.4% over the state‐of‐the‐art baseline method, demonstrating its effectiveness and efficiency. The codes used are publicly available at https://github.com/YeXinD/Rotation‐Invariant‐Attack.

In multi-layer heterogeneous satellite internet, the time-varying topology, dynamic channel quality, unstable satellites and emerging high-dimensional data present daunting challenges to its maintenance and management system.Thanks to its ability to learn from massive historical data or interaction with dynamic environment, artificial intelligence has been considered to be a promising solution to meet the aforementioned challenge.In this paper, we reviewed the research status of intelligent satellite internet maintenance and management technology on the following three aspects, network resource management, traff c fl ow management and network state management.Based on the above introduction, we further discussed the limitation of existing research and discussed promising future research direction of this fi eld as well.

Traffic prediction plays a significant part in creating intelligent cities such as traffic management, urban computing, and public safety. Nevertheless, the complex spatio-temporal linkages and dynamically shifting patterns make it somewhat challenging. Existing mainstream traffic prediction approaches heavily rely on graph convolutional networks and sequence prediction methods to extract complicated spatio-temporal patterns statically. However, they neglect to account for dynamic underlying correlations and thus fail to produce satisfactory prediction results. Therefore, we propose a novel Self-Adaptive Spatio-Temporal Graph Convolutional Network (SASTGCN) for traffic prediction. A self-adaptive calibrator, a spatio-temporal feature extractor, and a predictor comprise the bulk of the framework. To extract the distribution bias of the input in the self-adaptive calibrator, we employ a self-supervisor made of an encoder–decoder structure. The concatenation of the bias and the original characteristics are provided as input to the spatio-temporal feature extractor, which leverages a transformer and graph convolution structures to learn the spatio-temporal pattern, and then applies a predictor to produce the final prediction. Extensive trials on two public traffic prediction datasets (METR-LA and PEMS-BAY) demonstrate that SASTGCN surpasses the most recent techniques in several metrics.

Accurate time series forecasting is of great importance in real-world scenarios such as health care, transportation, and finance. Because of the tendency, temporal variations, and periodicity of the time series data, there are complex and dynamic dependencies among its underlying features. In time series forecasting tasks, the features learned by a specific task at the current time step (such as predicting mortality) are related to the features of historical timesteps and the features of adjacent timesteps of related tasks (such as predicting fever). Therefore, capturing dynamic dependencies in data is a challenging problem for learning accurate future prediction behavior. To address this challenge, we propose a cross-timestep feature-sharing multi-task time series forecasting model that can capture global and local dynamic dependencies in time series data. Initially, the global dynamic dependencies of features within each task are captured through a self-attention mechanism. Furthermore, an adaptive sparse graph structure is employed to capture the local dynamic dependencies inherent in the data, which can explicitly depict the correlation between features across timesteps and tasks. Lastly, the cross-timestep feature sharing between tasks is achieved through a graph attention mechanism, which strengthens the learning of shared features that are strongly correlated with a single task. It is beneficial for improving the generalization performance of the model. Our experimental results demonstrate that our method is significantly competitive compared to baseline methods.

Adversarial example generation on Synthetic Aperture Radar (SAR) images is an important research area that could have significant impacts on security and environmental monitoring. However, most current adversarial attack methods on SAR images are designed for white-box situations by end-to-end means, which are often difficult to achieve in real-world situations. This article proposes a novel black-box targeted attack method, called Shallow-Feature Attack (SFA). Specifically, SFA assumes that the shallow features of the model are more capable of reflecting spatial and semantic information such as target contours and textures in the image. The proposed SFA generates ghost data packages for input images and generates critical features by extracting gradients and feature maps at shallow layers of the model. The feature-level loss is then constructed using the critical features from both clean images and target images, which is combined with the end-to-end loss to form a hybrid loss function. By fitting the critical features of the input image at specific shallow layers of the neural network to the target critical features, our attack method generates more powerful and transferable adversarial examples. Experimental results show that the adversarial examples generated by the SFA attack method improved the success rate of single-model attack under a black-box scenario by an average of 3.73%, and 4.61% after combining them with ensemble-model attack without victim models.

The interpretation of optical and synthetic aperture radar (SAR) images in remote sensing is general for many tasks, such as environmental monitoring, marine management, and resource planning. Instance segmentation of optical and SAR images, which can simultaneously provide instance-level localization and pixel-level classification of objects of interest, is a crucial and challenging task in image interpretation. Considering that most current methods for instance segmentation of optical and SAR images rely on expensive pixel-level annotation, we develop a weakly supervised instance segmentation (WSIS) method to balance the visual processing requirements with the annotation cost. First, we decompose the prior knowledge of the mask-aware task in WSIS into three meta-knowledge components: fundamental knowledge, apparent knowledge, and detailed knowledge inspired by human visual perception habits of “whole to part” and “coarse to detailed.” Then, a meta-knowledge-guided weakly supervised instance segmentation network (MGWI-Net) is proposed. In this network, the weakly supervised mask (WSM) head can instantiate both fundamental knowledge and apparent knowledge to perform mask awareness without any annotations at the pixel level. The network also includes a mask information awareness assist (MIAA) head, which can implicitly guide the network to learn detailed information about edges through the boundary-sensitive feature of the fully connected conditional random field (CRF), facilitating the instantiation of detailed knowledge. The experimental results show that the MGWI-Net can efficiently generate instance masks for optical and SAR images and achieve the approximate instance segmentation results of the fully supervised method with about one-eighth of the annotation production time. The model parameters and processing speed of our network are also competitive. This study can provide inexpensive and convenient technical support for applying and promoting instance segmentation methods for optical and SAR images.

This paper proposes a network anomaly detection model of direct batch growing hierarchical self-organizing mapping based on entropy, which facilitates clear topology representation for the asymmetrically-distributed data. Since the entropy-defined parameters dynamically vary with the incident dataset, that is, follow a data-adaptive manner, the proposed model is naturally valid in all cases with various data types. For fine-grained data distinguishing, a resemble entropy parameter is proposed for the first time to our best knowledge. The experimental results validate that the proposed model achieves a more efficient network anomaly detection than the conventional models, especially for real-world applications with unexpected anomaly data updating.

Multi-view clustering aims to obtain the perfect clusters with a set of feature sets. Many methods learn a common agreement among views to achieve this. However, they may fail to use the specific unique feature of each view, which is important to describe the intrinsic characteristics of data. Unlike many works, we treat each view as a subset feature of data and fuse representations from those unique views to learn an integrated graph for clustering. We propose a novel representations fusion method for MVC. In this method, a regularized semi-nonnegative matrix factorization is proposed to learn the low-dimensional representation of each view, in which, a regularization term is designed to keep the neighboring structure in new low-dimensional space, and it is further updated and utilized to guide the learning process. Then the learned representations is fused to get an integrated graph for describing the image data. Finally, extensive experiments on several real image datasets show that our method achieves better performance than the state-of-the-art methods.

Applying an accurate parametric prediction model to identify abnormal or false pressurizer water levels (PWLs) is critical to the safe operation of marine pressurized water reactors (PWRs). Recently, deep-learning-based models have proved to be a powerful feature extractor to perform high-accuracy prediction. However, the effectiveness of models still suffers from two issues in PWL prediction: the correlations shifting over time between PWL and other feature parameters, and the example imbalance between fluctuation examples (minority) and stable examples (majority). To address these problems, we propose a cost-sensitive mechanism to facilitate the model to learn the feature representation of later examples and fluctuation examples. By weighting the standard mean square error loss with a cost-sensitive factor, we develop a Cost-Sensitive Long Short-Term Memory (CSLSTM) model to predict the PWL of PWRs. The overall performance of the CSLSTM is assessed by a variety of evaluation metrics with the experimental data collected from a marine PWR simulator. The comparisons with the Long Short-Term Memory (LSTM) model and the Support Vector Regression (SVR) model demonstrate the effectiveness of the CSLSTM.

Multi-view 3D reconstruction technology based on deep learning is developing rapidly. Unsupervised learning has become a research hotspot because it does not need ground truth labels. The current unsupervised method mainly uses 3DCNN to regularize the cost volume to regression image depth. This approach results in high memory requirements and long computing time. In this paper, we propose an end-to-end unsupervised multi-view 3D reconstruction network framework based on PatchMatch, Unsup_patchmatchnet. It dramatically reduces memory requirements and computing time. We propose a feature point consistency loss function. We incorporate various self-supervised signals such as photometric consistency loss and semantic consistency loss into the loss function. At the same time, we propose a high-resolution loss method. This improves the reconstruction of high-resolution images. The experiment proves that the memory usage of the network is reduced by 80% and the running time is reduced by more than 50% compared with the network using 3DCNN method. The overall error of reconstructed 3D point cloud is only 0.501 mm. It is superior to most current unsupervised multi-view 3D reconstruction networks. Then, we test on different data sets and verify that the network has good generalization.

Monocular 3D human pose estimation is used to calculate a 3D human pose from monocular images or videos. It still faces some challenges due to the lack of depth information. Traditional methods have tried to disambiguate it by building a pose dictionary or using temporal information, but these methods are too slow for real-time application. In this paper, we propose a real-time method named G2O-pose, which has a high running speed without affecting the accuracy so much. In our work, we regard the 3D human pose as a graph, and solve the problem by general graph optimization (G2O) under multiple constraints. The constraints are implemented by algorithms including 3D bone proportion recovery, human orientation classification and reverse joint correction and suppression. When the depth of the human body does not change much, our method outperforms the previous non-deep learning methods in terms of running speed, with only a slight decrease in accuracy.

With the rapid progress of urbanization, predicting citywide crowd flows has become increasingly significant in many fields, such as traffic management and public security. However, influenced by the complex spatiotemporal relations in raw data and other factors, such as events and weather, obtaining a precise prediction is challenging. Some previous works attempted to address this problem using various ways, such as autoregressive integrated moving average, vector auto-regression and some deep learning models. However, seldom can these methods comprehensively capture the spatiotemporal correlations. In this paper, we propose a novel spatio-temporal prediction model that is based on densely connected convolutional networks and attention long short-term memory (ST-DCCNAL), to simultaneously predict the inflow and outflow of the crowds in regions divided within a specific city. The ST-DCCNAL model consists of three parts: spatial part, external factors part and temporal part. In the spatial part, we employ densely connected convolutional networks to extract spatial characteristics at different levels. The external factors part utilizes a fully connected network to extract features from auxiliary information. In the last part, an attention-based long short-term memory module is leveraged to capture the temporal pattern. To demonstrate the practicality and effectiveness of the proposed model, we evaluate it using two separate real-world datasets of taxis in Beijing and bikes in New York. The experimental results confirm that the performance of our model is better than that of other baseline methods.

In foreground segmentation, it is challenging to construct an effective background model to learn the spatial-temporal representation of the background. Recently, deep learning-based background models (DBMs) with the capability of extracting high-level features have shown remarkable performance. However, the existing state-of-the-art DBMs deal with video segmentation as single-image segmentation and ignore temporal cues in video sequences. To exploit temporal data sufficiently, this paper proposes a multi-input multi-output (MIMO) DBM framework for the first time, which is partially inspired by the binocular summation effect in human eyes. Our framework is an X-shaped network which allows the DBM to track temporal changes in a video sequence. Moreover, each output branch of our model could receive visual signals from two similar input frames simultaneously like the binocular summation mechanism. In addition, our model can be trained end-to-end using only a few training examples without any post-processing. We evaluate our method on the largest dataset for change detection (CDnet 2014) and achieve the state-of-the-art performance by an average overall F-Measure of 0.9920.

Although person re-identification (ReID) has drawn increasing research attention due to its potential to address the problem of analysis and processing of massive monitoring data, it is very challenging to learn discriminative information when the people in the images are occluded, in large pose variations or from different perspectives. To address this problem, we propose a novel joint attention person ReID (JA-ReID) architecture. The idea is to learn two complementary feature representations by combining a soft pixel-level attention mechanism and a hard region-level attention mechanism. The soft pixel-level attention mechanism learns a discriminative embedding for the fine-grained information by exploring the salient parts in the feature maps. The hard region-level attention mechanism conducts uniform partitions on the convolutional feature maps for learning local features. We have achieved competitive results in three popular benchmarks, including Market1501, DukeMTMC-reID, and CUHK03. The experimental results verify the adaptability of the joint attention mechanism to non-rigid deformation of the human body, which can effectively improve the accuracy of ReID.

Malware and its variants continue to pose a threat to network security. Machine learning has been widely used in the field of malware classification, but some emerging studies, such as attention mechanisms, are rarely applied in this field. In this paper, we analyze the correspondence between bytecode and disassembly of malware, and propose a new feature extraction method based on multi-dimensional sequence. Also, we construct a new classification framework based on attention mechanism and Convolutional Neural Networks mechanism. Furthermore, we also compare the different architectures based on the attention mechanisms. Experiments on open datasets show that our feature extraction method and our framework have a good classification effect, and the accuracy rate is 0.9609.

Video compression sensing can use a few measurements to obtain the original video by reconstruction algorithms. There is a natural correlation between video frames, and how to exploit this feature becomes the key to improving the reconstruction quality. More and more deep learning-based video compression sensing (VCS) methods are proposed. Some methods overlook interframe information, so they fail to achieve satisfactory reconstruction quality. Some use complex network structures to exploit the interframe information, but it increases the parameters and makes the training process more complicated. To overcome the limitations of existing VCS methods, we propose an efficient end-to-end VCS network, which integrates the measurement and reconstruction into one whole framework. In the measurement part, we train a measurement matrix rather than a pre-prepared random matrix, which fits the video reconstruction task better. An unfolded LSTM network is utilized in the reconstruction part, deeply fusing the intra- and interframe spatial–temporal information. The proposed method has higher reconstruction accuracy than existing video compression sensing networks and even performs well at measurement ratios as low as 0.01.

Current distillation methods only distill between corresponding layers, and do not consider the knowledge contained in preceding layers. To solve this problem, we analyzed the guiding effect of the inferior features of a teacher model on the coordinate feature of a student model, and proposed inferior and coordinate distillation for object detectors. The proposed method utilizes the rich information contained in different layers of the teacher model; such that the student model can review the old information and learn the new information, in addition to the dark knowledge in the teacher model. Moreover, the refine module is used to align the features of different layers, distinguish the spatial and channel to extract attention, strengthen the correlation between the features of different stages, and prevent the disorder caused by merging. Exclusive experiments were conducted on different object detectors. The results for the mean average precision (mAP) obtained using Faster R-CNN, RetinaNet, and fully convolutional one-stage object detector (FCOS) with ResNet-50 as its backbone were 40.5%, 39.8%, and 42.8% with regard to the COCO dataset, respectively; which are 2.1%, 2.4%, and 4.3% higher than the benchmark, respectively.

Large-scale terminals’ various QoS requirements are key challenges confronting the resource allocation of Satellite Internet of Things (S-IoT). This paper presents a deep reinforcement learning-based online channel allocation and power control algorithm in an S-IoT uplink scenario. The intelligent agent determines the transmission channel and power simultaneously based on contextual information. Furthermore, the weighted normalized reward concerning success rate, power efficiency, and QoS requirement is adopted to balance the performance between increasing resource efficiency and meeting QoS requirements. Finally, a practical deployment mechanism based on transfer learning is proposed to promote onboard training efficiency and to reduce computation consumption of the training process. The simulation demonstrates that the proposed method can balance the success rate and power efficiency with QoS requirement guaranteed. For S-IoT’s normal operation condition, the proposed method can improve the power efficiency by 60.91% and 144.44% compared with GA and DRL_RA, while its power efficiency is only 4.55% lower than that of DRL-EERA. In addition, this method can be transferred and deployed to a space environment by merely 100 onboard training steps.

In this paper, we propose a multi-scene adaptive crowd counting method based on meta-knowledge and multi-task learning. In practice, surveillance cameras are stationarily deployed in various scenes. Considering the extensibility of a surveillance system, the ideal crowd counting method should have a strong generalization capability to be deployed in unknown scenes. On the other hand, given the diversity of scenes, it should also effectively suit each scene for better performance. These two objectives are contradictory, so we propose a coarse-to-fine pipeline including meta-knowledge network and multi-task learning. Specifically, at the coarse-grained stage, we propose a generic two-stream network for all existing scenes to encode meta-knowledge especially inter-frame temporal knowledge. At the fine-grained stage, the regression of the crowd density map to the overall number of people in each scene is considered a homogeneous subtask in a multi-task framework. A robust multi-task learning algorithm is applied to effectively learn scene-specific regression parameters for existing and new scenes, which further improve the accuracy of each specific scenes. Taking advantage of multi-task learning, the proposed method can be deployed to multiple new scenes without duplicated model training. Compared with two representative methods, namely AMSNet and MAML-counting, the proposed method reduces the MAE by 10.29% and 13.48%, respectively.

Area Under the ROC Curve (AUC) is a widely used metric for measuring classification performance. It has important theoretical and academic values to develop AUC maximization algorithms. Traditional methods often apply batch learning algorithm to maximize AUC which is inefficient and unscalable for large-scale applications. Recently some online learning algorithms have been introduced to maximize AUC by going through the data only once. However, these methods sometimes fail to converge to an optimal solution due to the fixed or rapid decay of learning rates. To tackle this problem, we propose an algorithm AdmOAM, Adaptive Moment estimation method for Online AUC Maximization. It applies the estimation of moments of gradients to accelerate the convergence and mitigates the rapid decay of the learning rates. We establish the regret bound of the proposed algorithm and implement extensive experiments to demonstrate its effectiveness and efficiency.

Identifying influential nodes is an important topic in many diverse applications, such as accelerating information propagation, controlling rumors and diseases. Many methods have been put forward to identify influential nodes in complex networks, ranging from node centrality to diffusion-based processes. However, most of the previous studies do not take into account overlapping communities in networks. In this paper, we propose an effective method based on network representation learning. The method considers not only the overlapping communities in networks, but also the network structure. Experiments on real-world networks show that the proposed method outperforms many benchmark algorithms and can be used in large-scale networks.

Community detection is an important tasks across a number of research fields including social science, biology, and physics. In the real world, topology information alone is often inadequate to accurately find out community structure due to its sparsity and noise. The potential useful prior information such as pairwise constraints which contain must-link and cannot-link constraints can be obtained from domain knowledge in many applications. Thus, combining network topology with prior information to improve the community detection accuracy is promising. Previous methods mainly utilize the must-link constraints while cannot make full use of cannot-link constraints. In this paper, we propose a semi-supervised community detection framework which can effectively incorporate two types of pairwise constraints into the detection process. Particularly, must-link and cannot-link constraints are represented as positive and negative links, and we encode them by adding different graph regularization terms to penalize closeness of the nodes. Experiments on multiple real-world datasets show that the proposed framework significantly improves the accuracy of community detection.