Bingxian Wu, Yu Zhang, Zonghao Guo, Tang Liu, Chen Qian, Yuxiang Lu, Xingbo Du, Yanghao Li, Yidan Zhang, Chi Chen, Ling Yao, Chenghu Zhou, Maosong Sun. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Rapid development of web services has led to an explosion of multimodal content, making multimodal recommender systems (MRSs) vital tools for mitigating information overload. Current MRSs have achieved remarkable progress by incorporating advanced technologies such as Graph Neural Networks (GNNs) and Large Language Models (LLMs). However, these studies still suffer from the semantic shift problem. Generally, item's multimodal content usually contain multiple objects, including target object (core content of item) and auxiliary objects (decorations of item). Existing MRSs overlooked this distinction, failing to prevent auxiliary objects from dominating the representation, leading to biased item representation. To address this issue, we propose a model-agnostic framework ''TargetMR''. Concretely, TargetMR comprises two core modules, including Object Disentangler and Object Identifier. The Object Disentangler decouples item text and image into multiple objects via text syntactic parsing and image segmentation. The Object Identifier performs knowledge distillation based on LLMs to efficiently identify the target text object. It then identifies the target image object through cross-modal semantic evaluation. Moreover, this module refines the representation of image target object by optimizing the semantic correlation. Owing to the model-agnostic design of TargetMR, it can be integrated into various backbone MRSs. Extensive experiments on three benchmark datasets show that TargetMR consistently improves the performance of five backbone MRSs, with an average improvement of 12.26%. Our codes are available at https://github.com/gutang-97/TargetMR/.

Deep learning has demonstrated remarkable generalization capability with independent and identically distributed (i.i.d.) training and test data, however, it often struggles with data drawn from different, albeit causally related, distributions. This problem is generally known as Out-of-Distribution (OoD) generalization. While there is a plethora of algorithms proposed for OoD generalization, the current understanding of the data commonly employed to evaluate these algorithms remains relatively naive. In this study, we identify two distinct types of distribution shifts, namely diversity shift and correlation shift, that are ubiquitous in various OoD datasets. We propose a quantifiable formal definition for the two shifts and show that the performance of OoD algorithms is upper bounded by them. To validate our theoretical insight, we evaluate a number of OoD generalization algorithms across two groups of datasets from both classification and object detection areas, each dominated by one of the shifts, exposing the strengths of the algorithms against one shift as well as their limitations against the other. We further proved that all performance degradations according to data distribution shifts can be attributed to these two types of shifts defined in our paper. The benchmark integrates existing datasets and algorithms from different research areas that seem unrelated into a coherent picture, which may serve as a foundation for future OoD generalization research.

In real-world scenarios, training and test data are often collected in diverse settings, leading to domain shifts arising from evolving environments and selection bias. While causality-inspired methods have shown promising results in tackling the out-of-distribution (OOD) generalization issue, prior methods treat the discovered differences across domains as confounding variables. While effective in handling domain differences (i.e., unseen environmental features in test data), they may fail when confronted with intricate spurious correlations in real-world datasets. In this study, we first analyze this limitation to inadequate modeling of causal intervention and derive the OOD generalization bound to explain the challenges it introduces. To address this problem, we propose a modified causal intervention approach to mitigate various types of confounders. Motivated by the mathematical formulation of our modified causal intervention, we introduce the Causal Feature Selection Module (CFSM) to suppress model weights on both domain-differences features and spurious correlation features. Integrated within the Base Feature Extraction Module, In-Sample Module, and Cross-Sample Module (B-I-C architecture), CFSM collectively neutralizes the confounding effects arising from both domain discrepancies and correlation distinctions, thereby achieving causal feature selection. Under mild assumptions, we prove that the proposed CFSM method can achieve strictly lower OOD errors. Further experiments conducted on various benchmark datasets demonstrate the effectiveness of the proposed method. Compared to previous deconfounding methods, our method not only mitigates the effect of domain-differences features but also the hard-to-identify spurious correlation features, achieving significant improvements in two-dimensional OOD generalization.

Out-of-Distribution (OOD) generalization is a promising yet challenging goal that guarantees the test performance of Graph Neural Networks (GNNs) in open-world settings. However, due to the intricate internal topology of graph-structured data, redundant information from the spurious topologies severely confuses GNNs to deviate from the labels. Extracting concise and label-relevant subgraphs from the original graphs can alleviate this problem. Unfortunately, existing methods either overlook the global structural distribution or rely heavily on manually predefined assumptions. As a result, they fall short of well capturing the structural distribution changes between input graph and extracted subgraph, thus compromising adaptability of extracted invariant subgraphs to diverse OOD scenarios. This motivates us to propose a framework called S tructural- E ntropy-guided I nformation B ottleneck (OOD-SEIB) that aims to more traceably measure the inherent information changes for better and more flexible OOD generalization. The core of OOD-SEIB lies in concise topology extraction module, where we measure the mutual information flow between input graph and extracted subgraph based on structural entropy, termed Compression Index (CI). Specifically, the CI is a quantifiable metric that calculates the codeword length required to describe entire graph structure via a biased random walk. Under this guidance, OOD-SEIB then launches a structural information bottleneck compression module that jointly optimizes both CI and label-relevance of the subgraph topology by iteratively balancing between informativeness and compression. To further improve GNN’s invariant subgraph identification capability, OOD-SEIB generates multiple augmented environments and distill the invariant subgraphs into GNN as knowledge in an inside-out manner. When iteratively optimizing in above prescribed way, OOD-SEIB progressively reinforce the invariant subgraph extraction, thereby enhancing its generalization capability. Extensive experiments on synthetic and three real-world graph-level OOD benchmarks demonstrate that our proposed OOD-SEIB improves classification accuracy by 4.85%–38.03% on average compared to state-of-the-art baselines. Additionally, we extend OOD-SEIB to two node-level benchmarks, achieving average classification accuracy improvements of 14.52% and 13.15%.

The deployment of sensors enables data-driven urban management, but necessitates inductive spatio-temporal kriging to infer unmonitored areas. Existing methods impute these unknown observations by smoothing temporal features based on spatial dependencies, overlooking the decoupling ofinherent propertiesanddynamic correlationsin message passing. In particular, the inherent properties reveal non-transitive signals, and current coupled aggregation leads to inaccurate results. To this end, we proposeTempoRAl deCoupledKriging, named TRACK, to decouple two factors with the help of node-specific inherency. Specifically, we first construct a node-specific profile to represent its inherency including geographical and periodic features, which is subsequently transformed into decoupling prompts. Secondly, the coupled temporal features are separated through querying each prompt embedding, facilitating precise temporal aggregation for inherent properties and spatial aggregation for dynamic correlations. Finally, a multi-task training strategy is further adopted to mimic the inductive scenarios during testing. We evaluate TRACK on four real-world datasets spanning urban traffic and air quality prediction tasks. TRACK achieves state-of-the-art performance, with average improvements of 3.10% in MAE and 4.45% in RMSE over strong baselines. Moreover, we further demonstrated its robust generalization in a challenging cross-city inductive setting. Code is available athttps://github.com/JeremyChou28/TRACK.

This paper proposes a novel address mapping mechanism for multi-robot communication and collaboration systems, named SPing. SPing addresses each robot with a dynamicphysical-world address- an encoding of the robot's physical location - rather than a pre-assigned digital-world ID (e.g., the IP address). This enables a “see-and-point” communication mode for robots: a robot can establish an immediate connection pointing to any other robot it intends to collaborate with in its visual field, without relying on a pre-existing multi-robot network. This on one hand improves the robustness and usefulness of multi-robot systems in uncertain and unstructured environments where network infrastructures are unavailable. On the other hand, it makes the robots' communication behavior tightly coupled with and more supportive of the collaboration tasks in the physical world. We build an end-to-end prototype of SPing and evaluate its performance in both static and mobile scenarios. The results show that SPing can always establish a connection precisely pointing to the target receiver with an average matching rate of 99.58%, and a spatial resolution of 0.3 m$\sim$0.5m.

This paper tries to answer a question: “Can we achieve spatial-selective transmission on IoT devices?” A positive answer would enable more secure data transmission among IoT devices. The challenge, however, is how to manipulate signal propagation without relying on beamforming antenna arrays which are usually unavailable on low-end IoT devices. We give an affirmative answer by introducing SpotSound, a novel acoustic communication system that exploits the diversity of multi-path indoors as a naturalbeamformer. By judiciously controlling the way how the information is embedded into the signal, SpotSound can make the signal decodable only when the signal propagates along a certain multipath channel. Since the multipath channel decorrelates rapidly over the distance between receivers, SpotSound can ensure the signal is decodable only at the target position, achieving precise physical isolation. SpotSound is a purely software-based solution that can run on most IoT devices where speakers and microphones are widely used. We implement SpotSound on Raspberry Pi connected with COTS microphone and speaker. Experimental results show that SpotSound could precisely focus its signal on spots with customized sizes ranging from 0.04m2to 0.5m2.

This paper presents μTag, an ultra-low-power backscatter sensor that supports high-rate sensing of a large number of targets simultaneously. The core of μTag is an RF “gene editing” technique that embeds both the identity of the sensor and the real-time motion or vibration state of the attached target intensively in the transient features of the sensor’s RF signal, in a collision-resilient manner. We provide practical techniques which i) generate such “genetic signal” with purely analog and extremely simple circuits; and ii) separate the signals from a large scale of sensors reliably. Our experimental results show that our design can support concurrent tracking of 150 targets with a 12 kHz per-tag sampling rate. We also demonstrate with multiple sensing applications that μTag can achieve high-speed and large-scale motion tracking, rotation frequency sensing, and contactless sensing. The PCB power consumption of μTag is 38∼107 μW, according to the operating frequency of the tag. Our ASIC simulation based on the 40 nm CMOS process shows that the power consumption can be further reduced to 0.13∼0.52 μW.

Transformer-based models have achieved great success in document ranking, yet they suffer from substantial computational costs due to the quadratic complexity of attention, particularly for Long Document Ranking (LDR). Token pruning is a promising approach to reducing computational costs, while existing methods have largely overlooked the interaction and alignment between document and query for guiding the token pruning process, which may result in mistakenly pruned tokens that are critical for the query but not for the document. Additionally, these methods often lack the flexibility required for varying input samples. To this end, we propose a novel framework, Adaptive Token Pruning with Query-Document Alignment (QD-ATP) for accelerating LDR. Specifically, we first introduce a well-designed Query-Document Alignment Guidance (QDAG) module to effectively align the semantic and matching information between query and document, ensuring that the pruned tokens are less important for both two fields. Furthermore, we design a novel Adaptive Token Pruning (ATP) module, which can dynamically adjust pruning ratios based on different input samples. Experimental results on three benchmark datasets demonstrate that QD-ATP can achieve up to a 7.3× latency speedup while preserving competitive performance.

With the development of data-centric AI, the focus has shifted from model-driven approaches to improving data quality. Academic literature, as one of the crucial types, is predominantly stored in PDF formats and needs to be parsed into texts before further processing. However, parsing diverse structured texts in academic literature remains challenging due to the lack of datasets that cover various text structures. In this paper, we introduce AceParse, the first comprehensive dataset designed to support the parsing of a wide range of structured texts, including formulas, tables, lists, algorithms, and sentences with embedded mathematical expressions. Based on AceParse, we fine-tuned a multimodal model, named AceParser, which accurately parses various structured texts within academic literature. This model outperforms the previous state-of-the-art by 4.1% in terms of F1 score and by 5% in Jaccard Similarity, demonstrating the potential of multimodal models in academic literature parsing. Our dataset is available at https://github.com/JHW5981/AceParse.

Evolving graphs are prevalent in real-world systems where nodes and edges continuously expand over time, resulting in persistent topological changes across temporal snapshots. As the graph expands, the increasing structural discrepancy between earlier and later timesteps often leads to poor generalization-a phenomenon widely recognized as Temporal out-of-distribution (TOOD) generalization. This issue has spurred growing research interest in generalized graph learning. Existing approaches often rely on modeling latent environments or incorporating structural causal reasoning, but they frequently face scalability limitations when applied to rapidly expanding graphs. In this work, we carry out a detailed analysis of temporal drift in evolving graphs and identify two complementary forms: structural role drift reflected by changes in degree proportions, and semantic drift measured by embedding divergence. To capture this graph evolution, we propose PromptEG, a scalable prompt-based framework that generates node-specific prompts using degree signals and KL-divergence between representations. These prompts are optimized via consistency and contrastive objectives, and injected into GNNs to enhance temporal generalization. Experiments on largescale evolving graphs-containing up to 500 K edges and tens of thousands of nodes-demonstrate that PromptEG achieves strong long-horizon generalization while maintaining high computational efficiency.

Given a style-reference image as the additional image condition, text-to-image diffusion models have demonstrated impressive capabilities in generating images that possess the content of text prompts while adopting the visual style of the reference image. However, current state-of-the-art methods often struggle to disentangle content and style from style-reference images, leading to issues such as content leakages. To address this issue, we propose a masking-based method that efficiently decouples content from style without the need of tuning any model parameters. By simply masking specific elements in the style reference's image features, we uncover a critical yet under-explored principle: guiding with appropriately-selected fewer conditions (e.g., dropping several image feature elements) can efficiently avoid unwanted content flowing into the diffusion models, enhancing the style transfer performances of text-to-image diffusion models. In this paper, we validate this finding both theoretically and experimentally. Extensive experiments across various styles demonstrate the effectiveness of our masking-based method and support our theoretical results.

Improving the generalization of multi-camera 3D object detection is essential for safe autonomous driving in the real world. In this paper, we consider a realistic yet more challenging scenario, which aims to improve the generalization when only single source data available for training, as gathering diverse domains of data and collecting annotations is time-consuming and labor-intensive. To this end, we propose the Fourier Cross-View Learning (FCVL) framework including Fourier Hierarchical Augmentation (FHiAug), an augmentation strategy in the frequency domain to boost domain diversity, and Fourier Cross-View Semantic Consistency Loss to facilitate the model to learn more domain-invariant features from adjacent perspectives. Furthermore, we provide theoretical guarantees via augmentation graph theory. To the best of our knowledge, this is the first study to explore generalizable multi-camera 3D object detection with a single source. Extensive experiments on various testing domains have demonstrated that our approach achieves the best performance across various domain generalization methods.

With the increasing popularity of autonomous driving based on the Bird's-Eye-View (BEV) representation, improving the generalization of such detection models is key for safe real-world applications. However, a realistic yet challenging scenario: Single Domain Generalization (SDG) for BEV, is still under-explored. A key ingredient for SDG is to increase data diversity via common image augmentation or adversarial data generation first. However, common image-level augmentation is not sufficient enough to ensure domain diversity in most part of latent space. The adversarial generation has the problem of unstable training or mode collapsing as well. To address these limitations, we present Tri-level Automatic Augmentation (Tri-AutoAug), a simple yet effective method to enlarge the diversity and quantity of data from image and 2D features and facilitate the model to learn more domain-invariant features in BEV space. Besides, Tri-AutoAug can automatically learn augmentation strategies to avoid spending too much time manually adjusting hyperparameters and maximize the benefit of Tri-level Augmentation. To the best of our knowledge, this is the first study to explore automatic augmentation for SDG BEV. Extensive experiments on NuScenes-C including eight testing domains have demonstrated that our approach can achieve the best performance across various domain generalization methods. More importantly, we evaluate the proposed method in real-world autonomous driving scenarios. Tri-AutoAug improves the out-of-distribution (ood) performance by 8.54% (mAP), which demonstrates that Tri-AutoAug provides a practical and feasible solution for the applications of 3D detectors in the real world. The code is available at https://github.com/ClaireTunlTri-AutoAug.

With the explosive growth of online multimodal content, multimodal recommender systems(MRSs) have brought significant benefits to multimedia platforms. As MRSs evolve, many studies incorporate advanced technologies like graph neural networks(GNNs) and self-supervised learning(SSL), achieving remarkable results. However, these efforts still suffer from the quality disparity problem. It refers to the mixture of high and low quality across items' multiple modalities, owing to disparities in construction costs or design levels. These low-quality modalities often lack crucial details or introduce noise to the depiction of item, leading to insufficient or polluted item representation. Therefore, we propose a novel framework R2MR: Review and Rewrite Modality for Recommendation to tackle this issue. Specifically, R2MR is composed of two key components: Modality Reviewer and Modality Rewriter. The Modality Reviewer introduces a Consensus Review Mechanism. It performs perspective decomposition based on user representations and learns the consensus quality scores for modalities from diverse perspectives of multiple users. The Modality Rewriter proposes a Latent Mapping Model, which improves the quality of inferior modalities by learning various mapping patterns from high-quality modalities. Comprehensive experiments across three benchmark datasets reveal that R2MR substantially outperforms state-of-the-art methods, achieving an average improvement of 9.20%. The implementations are available at https://github.com/gutang-97/R2MR.

Nowadays, the explosive growth of academic literature has been going far beyond scientists' limited capability to read through, making it increasingly difficult for them to absorb disciplinary insights and extract intellectual essences critical for generating novel research ideas in interdisciplinary studies. To address this, we develop DeepReport, an AI-assisted scientific idea generation system to alleviate the research burden. Technically, DeepReport maintains evolving concept co-occurrence graphs to extract core insights from over 260 million publications across all disciplines. These concepts are periodically collected and updated, enabling the automatic extraction of hidden cross-domain connections. Combining temporal link prediction and analysis techniques with large language models, DeepReport is able to further transform these patterns of insights into actionable ideas. With the function of integrating up-to-date academic databases, visualizing dynamic relationships of concepts, and automatically generating new ideas, DeepReport empowers researchers to navigate complex knowledge landscapes, reduce cognitive burdens, and accelerate the generation of groundbreaking concepts. This work provides an in-depth exploration of DeepReport's architecture, functionalities, and applications, highlighting its transformative potential for advancing interdisciplinary research and fostering innovation. DeepReport is available at https://idea.acemap.cn/.

As the domain of artificial intelligence has advanced, the integration of deep learning techniques into terrain and landform analysis has become more prevalent. Nevertheless, many existing methods are fully supervised and designed for specific tasks; thus, their transferability is limited and massive annotated samples are required. This study introduces a geomorphic pretrained model (GeomorPM) capable of performing multiple tasks. First, an architecture was designed that combined a convolution-based Vector Quantised-Variational Autoencoder (VQVAE) with a Transformer-based masked autoencoder (MAE) framework, allowing it to autonomously learn local details and global patterns from large-scale digital elevation model (DEM) data. Subsequently, GeomorPM, based on the VQMAE architecture, was pretrained on massive DEM data and fine-tuned for three specific tasks: DEM void filling, DEM superresolution, and landform classification. GeomorPM outperformed the traditional and other deep learning methods in all three tasks, demonstrating the superior learning ability and transferability of the model. This study provides a practical framework for developing pretrained models based on DEMs that can be expanded to other continuous geoscientific data.

Constructing spatial interaction (SI) networks from massive geographic flow data is an important method for understanding the correlation patterns and trends among regions. However, large-scale SI networks usually cause complex edge interleaving and a sharp increase in noise information. A network skeleton can represent the key structure of the corresponding complex network with generalized network data. However, the existing extraction methods only consider the topological information of networks and ignore spatial information, which may cause difficulty maintaining the spatial distribution characteristics and the exclusion of key spatial information. To address the issue, we have developed a skeleton extraction method for large-scale SI networks. First, the network nodes are divided into different clusters via adaptive spatial clustering. Second, an entropy weight method is used to define a comprehensive evaluation indicator of the network node importance, and the nodes in each cluster are sorted and selected according to the comprehensive evaluation indicator. Finally, the skeleton structure of the SI network is reconstructed according to the selected nodes and their adjacent edges. Compared with the existing methods for extracting network skeletons, the proposed method can maintain the spatial distribution characteristics of SI networks while preserving key topological features.

This paper delves into the challenge of maintaining connectivity in adversarial networks, focusing on the preservation of essential links to prevent the disintegration of network components under attack. Unlike previous approaches that assume a stable and homogeneous network topology, this study introduces a more realistic model that incorporates both link uncertainty and heterogeneity. Link uncertainty necessitates additional probing to confirm link existence, while heterogeneity reflects the varying resilience of links against attacks. We model the network as a random graph where each link is defined by its existence probability, probing cost, and resilience. The primary objective is to devise a defensive strategy that maximizes the expected size of the largest connected component at the end of an adversarial process while minimizing the probing cost, irrespective of the attack patterns employed. We begin by establishing the NP-hardness of the problem and then introduce an optimal defensive strategy based on dynamic programming. Due to the high computational cost of achieving optimality, we also develop two approximate strategies that offer efficient solutions within polynomial time. The first is a heuristic method that assesses link importance across three heterogeneous subnetworks, and the second is an adaptive minimax policy designed to minimize the defender’s potential worst-case loss, with guaranteed performance. Through extensive testing on both synthetic and real-world datasets across various attack scenarios, our strategies demonstrate significant advantages over existing methods.

In real-world scenarios, distribution shifts give rise to the importance of two problems: out-of-distribution (OoD) generalization, which focuses on models' generalization ability against covariate shifts (i.e., the changes of environments), and OoD detection, which aims to be aware of semantic shifts (i.e., test-time unseen classes). Real-world testing environments often involve a combination of both covariate and semantic shifts. While numerous methods have been proposed to address these critical issues, only a few works tackled them simultaneously. Moreover, prior works often improve one problem but sacrifice the other. To overcome these limitations, we delve into boosting OoD detection and OoD generalization from the perspective of information theory, which can be easily applied to existing models and different tasks. Building upon the theoretical bounds for mutual information and conditional entropy, we provide a unified approach, composed of Mutual Information Minimization (MI-Min) and Conditional Entropy Maximizing (CE-Max). Extensive experiments and comprehensive evaluations on multi-label image classification and object detection have demonstrated the superiority of our method. It successfully mitigates trade-offs between the two challenges compared to competitive baselines.

Accurate precipitation nowcasting up to two hours in advance is crucial for weather-sensitive decision-making, such as flood disaster warnings. Although deep learning-based radar echo extrapolation methods show advantages, they still face challenges, particularly in predicting local spatial variations and mitigating performance degradation with increasing prediction time horizon. This study proposes a new multi-scale spatiotemporal prediction LSTM model (MS2T-LSTM) to enhance precipitation nowcasting capabilities by fusing multi-scale spatial features through parallel convolutional structures and improving spatiotemporal reasoning through stacked LSTMs. Notably, multi-scale spatiotemporal feature gate is introduced into LSTM unit to transmit multi-scale spatiotemporal features of radar echo image sequences, mitigating the issue of time-dependent performance degradation. The loss function for model optimization is carefully selected via series of comparative experiments. The experiments show that MS2T-LSTM outperforms benchmark models in predicting precipitation events with different intensities. Specifically, in comparison to PredRNN++, the F1 score, Critical Success Index, Heidke Skill Score, and Probability of Detection for low-intensity precipitation threshold (5mm/h) improved by 2.3%, 3.5%, 13%, and 10.6%. This study demonstrates the effectiveness of integrating multi-scale spatiotemporal features to enhance precipitation forecasting, providing valuable insights for spatiotemporal prediction modeling and optimization.

Due to detector malfunctions and communication failures, missing data is ubiquitous during the collection of traffic data. Therefore, it is of vital importance to impute the missing values to facilitate data analysis and decision-making for Intelligent Transportation System (ITS) . However, existing imputation methods generally perform zero pre-filling techniques to initialize missing values, introducing inevitable noise. Moreover, we observe prevalent over-smoothed interpolations, falling short in revealing the intrinsic spatio-temporal correlations of incomplete traffic data. To this end, we propose Mask-Aware Graph Imputation Network (MagiNet) . Our method designs an adaptive mask spatio-temporal encoder to learn the latent representations of incomplete data, eliminating the reliance on pre-filling missing values. Furthermore, we devise a spatio-temporal decoder that stacks multiple blocks to capture the inherent spatial and temporal dependencies within incomplete traffic data, alleviating over-smoothed imputation. Extensive experiments demonstrate that our method outperforms state-of-the-art imputation methods on five real-world traffic datasets, yielding an average improvement of 4.31% in RMSE and 3.72% in MAPE under Missing Completely at Random (MCAR) pattern. Code is available at https://github.com/JeremyChou28/MagiNet .

Node importance estimation (NIE) is the task of inferring the importance scores of the nodes in a graph. Due to the availability of richer data and knowledge, recent research interests of NIE have been dedicated to knowledge graphs (KGs) for predicting future or missing node importance scores. Existing state-of-the-art NIE methods train the model by available labels, and they consider every interested node equally before training. However, the nodes with higher importance often require or receive more attention in real-world scenarios, e.g., people may care more about the movies or webpages with higher importance. To this end, we introduce Label Informed ContrAstive Pretraining (LICAP) to the NIE problem for being better aware of the nodes with high importance scores. Specifically, LICAP is a novel type of contrastive learning (CL) framework that aims to fully utilize continuous labels to generate contrastive samples for pretraining embeddings. Considering the NIE problem, LICAP adopts a novel sampling strategy called top nodes preferred hierarchical sampling to first group all interested nodes into a top bin and a nontop bin based on node importance scores, and then divide the nodes within the top bin into several finer bins also based on the scores. The contrastive samples are generated from those bins and are then used to pretrain node embeddings of KGs via a newly proposed predicate-aware graph attention networks (PreGATs), so as to better separate the top nodes from nontop nodes, and distinguish the top nodes within the top bin by keeping the relative order among finer bins. Extensive experiments demonstrate that the LICAP pretrained embeddings can further boost the performance of existing NIE methods and achieve new state-of-the-art performance regarding both regression and ranking metrics. The source code for reproducibility is available at https://github.com/zhangtia16/LICAP.

We investigate tight end-to-end delay guarantees for real-time flows with stochastic arrivals in multi-hop networks. The existing multi-hop scheduling policy either only serves for real-time flows with deterministic arrivals or cannot offer a tight end-to-end delay guarantee for multi-hop flows. In this paper, we prove a closed-form formula, which offers a sufficient and necessary condition for the EDF (Earliest Deadline First) scheduling policy to meet all the end-to-end deadlines for flows with stochastic arrivals in converge-cast tree networks. To the best of our knowledge, this is the first formula that characterizes the exact schedulability region for EDF in multi-hop networks. Moving beyond converge-cast tree networks to general multi-hop networks, we introduce the Segment EDF approach. This method partitions a general network into multiple converge-cast networks using a novel concept called the critical links. By determining segment deadlines within the schedulability region of each converge-cast tree, Segment EDF offers a tight end-to-end delay guarantee for each flow. We present a theoretical analysis showcasing the superior performance of Segment EDF over the existing scheduling policy of hop-by-hop EDF. Furthermore, we evaluate the performance of Segment EDF based on two key metrics: guaranteed flow completion time and admission ratio, in both real-world and synthetic networks. Our simulation results show that Segment EDF can provide 61.31%-96.4% tighter end-to-end delay guarantee and increase admission ratio by about 3.52%-107.59% for sequential arrival flow set and 1.04%-95.50% for batch arrival flow set than hop-by-hop EDF.

In this paper, we focus on a realistic yet challenging task, Single Domain Generalization Object Detection (S-DGOD), where only one source domain's data can be used for training object detectors, but have to generalize multiple distinct target domains. In S-DGOD, both high-capacity fitting and generalization abilities are needed due to the task's complexity. Differentiable Neural Architecture Search (NAS) is known for its high capacity for complex data fitting and we propose to leverage Differentiable NAS to solve S-DGOD. However, it may confront severe over-fitting issues due to the feature imbalance phenomenon, where parameters optimized by gradient descent are biased to learn from the easy-to-learn features, which are usually non-causal and spuriously correlated to ground truth labels, such as the features of background in object detection data. Consequently, this leads to serious performance degradation, especially in generalizing to unseen target domains with huge domain gaps between the source domain and target domains. To address this issue, we propose the Generalizable loss (G-loss), which is an OoD-aware objective, preventing NAS from over-fitting by using gradient descent to optimize parameters not only on a subset of easy-to-learn features but also the remaining predictive features for generalization, and the overall framework is named G-NAS. Experimental results on the S-DGOD urban-scene datasets demonstrate that the proposed G-NAS achieves SOTA performance compared to baseline methods. Codes are available at https://github.com/wufan-cse/G-NAS.

Out-of-distribution (OOD) generalization has long been a challenging problem that remains largely unsolved. Gaussian processes (GP), as popular probabilistic model classes, especially in the small data regime, presume strong OOD generalization abilities. Surprisingly, their OOD generalization abilities have been under-explored before compared with other lines of GP research. In this paper, we identify that GP is not free from the problem and propose a domain invariant learning algorithm for Gaussian processes (DIL-GP) with a min-max optimization on the likelihood. DIL-GP discovers the heterogeneity in the data and forces invariance across partitioned subsets of data. We further extend the DIL-GP to improve Bayesian optimization's adaptability on changing environments. Numerical experiments demonstrate the superiority of DIL-GP for predictions on several synthetic and real-world datasets. We further demonstrate the effectiveness of the DIL-GP Bayesian optimization method on a PID parameters tuning experiment for a quadrotor. The full version and source code are available at: https://github.com/Billzxl/DIL-GP.

Missing values are prevalent in multivariate time series, compromising the integrity of analyses and degrading the performance of downstream tasks. Consequently, research has focused on multivariate time series imputation, aiming to accurately impute the missing values based on available observations. A key research question is how to ensure imputation consistency, i.e., intra-consistency between observed and imputed values, and inter-consistency between adjacent windows after imputation. However, previous methods rely solely on the inductive bias of the imputation targets to guide the learning process, ignoring imputation consistency and ultimately resulting in poor performance. Diffusion models, known for their powerful generative abilities, prefer to generate consistent results based on available observations. Therefore, we propose a conditional diffusion model for Multivariate Time Series Consistent Imputation (MTSCI). Specifically, MTSCI employs a contrastive complementary mask to generate dual views during the forward noising process. Then, the intra contrastive loss is calculated to ensure intra-consistency between the imputed and observed values. Meanwhile, MTSCI utilizes a mixup mechanism to incorporate conditional information from adjacent windows during the denoising process, facilitating the inter-consistency between imputed samples. Extensive experiments on multiple real-world datasets demonstrate that our method achieves the state-of-the-art performance on multivariate time series imputation task under different missing scenarios. Code is available at https://github.com/JeremyChou28/MTSCI.

The majority of GNNs are based on message-passing mechanisms. However, Message Passing Neural Networks (MPNNs) have inherent limitations in capturing long-range interactions. The exponentially growing node information is compressed into fixed-size representations through multiple rounds of message passing, leading to the over-squashing problem. This issue severely hinders the flow of information across the graph and creates a bottleneck in graph learning. The natural idea of introducing global attention to point-to-point communication, as adopted in Graph Transformers (GTs), lacks inductive biases on graph structures and relies on complex positional encodings to enhance their performance in practical tasks. In this paper, we observe that the sensitivity between nodes in MPNNs decreases exponentially with the shortest path distance. In contrast, GTs have constant sensitivity, which leads to a loss of inductive bias. To address these issues, we introduce structured state spaces to capture the hierarchy of rooted trees, achieving linear sensitivity with theoretical guarantees. We further propose a novel state-space model-based graph convolution, resulting in a new paradigm that retains both the strong inductive biases from MPNNs and the long-range modeling capabilities from GTs. Extensive experimental results on long-range and general graph benchmarks demonstrate the superiority of our approach.

Shuqian Sheng, Yi Xu, Tianhang Zhang, Zanwei Shen, Luoyi Fu, Jiaxin Ding, Lei Zhou, Xiaoying Gan, Xinbing Wang, Chenghu Zhou. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Human trajectories, representing sequences of location points over time, are extensively collected and analyzed for various real-world applications such as urban planning, transportation management, and personalized location-based services. Trajectory embedding transforms raw trajectories into vector representations, capturing the underlying patterns and structures in the data. However, the abstraction provided by vector representations introduces significant security and privacy risks. These embeddings, often shared between entities or organizations, can be exploited by adversaries to reconstruct original trajectories, thereby compromising individual privacy. In this paper, we investigate the privacy issues of trajectory embeddings from an adversary's perspective. We propose two types of attacks to reconstruct original trajectories using road network information, addressing scenarios where the adversary has varying degrees of access to the black-box representation model. The first attack assumes unrestricted access to the model, allowing the adversary to construct a large-scale dataset and train a neural network to predict the road sequence of the trajectories. The second attack considers limited access, where the adversary computes distance coordinates between selected trajectory landmarks and road segments to infer different parts of the trajectory. Our experiments on a real-world dataset demonstrate that the reconstructed trajectories outperform baseline methods, achieving substantially lower reconstruction errors and more accurate alignment with the original trajectories, highlighting the significant vulnerability of trajectory embeddings to privacy breaches. These findings underscore the need for robust privacy-preserving mechanisms in spatio-temporal data analysis.

The sheer volume of scientific experimental results and complex technical statements, often presented in tabular formats, presents a formidable barrier to individuals acquiring preferred information. The realms of scientific reasoning and content generation that adhere to user preferences encounter distinct challenges. In this work, we present a new task for generating fluent and logical descriptions that match user preferences over scientific tabular data, aiming to automate scientific document analysis. To facilitate research in this direction, we construct a new challenging dataset CTRLSciTab consisting of table-description pairs extracted from the scientific literature, with highlighted cells and corresponding domain-specific knowledge base. We evaluated popular pre-trained language models to establish a baseline and proposed a novel architecture outperforming competing approaches. The results showed that large models struggle to produce accurate content that aligns with user preferences. As the first of its kind, our work should motivate further research in scientific domains.1

Graph Neural Networks (GNNs) are widely deployed in vast fields, but they often struggle to maintain accurate representations as graphs evolve. We theoretically establish a lower bound, proving that under mild conditions, representation distortion inevitably occurs over time. To estimate the temporal distortion without human annotation after deployment, one naive approach is to pre-train a recurrent model (e.g., RNN) before deployment and use this model afterwards, but the estimation is far from satisfactory. In this paper, we analyze the representation distortion from an information theory perspective, and attribute it primarily to inaccurate feature extraction during evolution. Consequently, we introduce Smart, a straightforward and effective baseline enhanced by an adaptive feature extractor through self-supervised graph reconstruction. In synthetic random graphs, we further refine the former lower bound to show the inevitable distortion over time and empirically observe that Smart achieves good estimation performance. Moreover, we observe that Smart consistently shows outstanding generalization estimation on four real-world evolving graphs. The ablation studies underscore the necessity of graph reconstruction. For example, on OGB-arXiv dataset, the estimation metric MAPE deteriorates from 2.19% to 8.00% without reconstruction.

Graph pooling compresses graph information into a compact representation. State-of-the-art graph pooling methods follow a hierarchical approach, which reduces the graph size step-by-step. These methods must balance memory efficiency with preserving node information, depending on whether they use node dropping or node clustering. Additionally, fixed pooling ratios or numbers of pooling layers are predefined for all graphs, which prevents personalized pooling structures from being captured for each individual graph. In this work, inspired by bottom-up grammar induction, we propose an efficient graph parsing algorithm to infer the pooling structure, which then drives graph pooling. The resulting Graph Parsing Network (GPN) adaptively learns personalized pooling structure for each individual graph. GPN benefits from the discrete assignments generated by the graph parsing algorithm, allowing good memory efficiency while preserving node information intact. Experimental results on standard benchmarks demonstrate that GPN outperforms state-of-the-art graph pooling methods in graph classification tasks while being able to achieve competitive performance in node classification tasks. We also conduct a graph reconstruction task to show GPN's ability to preserve node information and measure both memory and time efficiency through relevant tests.

Accurately reconstructing the global ocean deoxygenation over a century is crucial for assessing and protecting marine ecosystem. Existing expert-dominated numerical simulations fail to catch up with the dynamic variation caused by global warming and human activities. Besides, due to the high-cost data collection, the historical observations are severely sparse, leading to big challenge for precise reconstruction. In this work, we propose OxyGenerator, the first deep learning based model, to reconstruct the global ocean deoxygenation from 1920 to 2023. Specifically, to address the heterogeneity across large temporal and spatial scales, we propose zoning-varying graph message-passing to capture the complex oceanographic correlations between missing values and sparse observations. Additionally, to further calibrate the uncertainty, we incorporate inductive bias from dissolved oxygen (DO) variations and chemical effects. Compared with in-situ DO observations, OxyGenerator significantly outperforms CMIP6 numerical simulations, reducing MAPE by 38.77%, demonstrating a promising potential to understand the “breathless ocean” in data-driven manner.

Recent vision-language pre-trained models (VL-PTMs) have shown remarkable success in open-vocabulary tasks. However, downstream use cases often involve further fine-tuning of VL-PTMs, which may distort their general knowledge and impair their ability to handle distribution shifts. In real-world scenarios, machine learning systems inevitably encounter both covariate shifts (e.g., changes in image styles) and semantic shifts (e.g., test-time unseen classes). This highlights the importance of enhancing out-of-distribution (OOD) generalization on covariate shifts and simultaneously detecting semantic-shifted unseen classes. Thus a critical but underexplored question arises: How to improve VL-PTMs’ generalization ability to closed-set OOD data, while effectively detecting open-set unseen classes during fine-tuning? In this paper, we propose a novel objective function of OOD detection that also serves to improve OOD generalization. We show that minimizing the gradient magnitude of energy scores on training data leads to domain-consistent Hessians of classification loss, a strong indicator for OOD generalization revealed by theoretical analysis. Based on this finding, we have developed a unified fine-tuning framework that allows for concurrent optimization of both tasks. Extensive experiments have demonstrated the superiority of our method. The code is available at https://github.com/LinLLLL/CRoFT.

This paper presents μTag, an ultra-low-power backscatter sensor that supports high-frequency sensing of a large number of targets simultaneously. The core of μTag is an RF "gene editing" technique that embeds both the identity of the sensor and the real-time motion state of the attached target intensively in the transient features of the sensor's RF signal, in a collision-resilient manner. We provide practical techniques which i) generate such "genetic signal" with purely analog and extremely simple circuits; and ii) separate the signals from a large scale of sensors reliably. Our experimental results show that our design can support concurrent tracking of 150 targets with a 12kHz per-tag sampling rate. We also demonstrate with multiple sensing applications that μTag can achieve high-speed and large-scale motion tracking and rotation frequency sensing. The PCB power consumption of μTag is 38~107μW, according to the operating frequency of the tag. Our ASIC simulation based on the 40nm CMOS process shows that the power consumption can be further reduced to 0.13~0.52μW.

Shuqian Sheng, Yi Xu, Luoyi Fu, Jiaxin Ding, Lei Zhou, Xinbing Wang, Chenghu Zhou. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

We investigate the entity alignment (EA) problem with unlabeled dangling cases, meaning that partial entities have no counterparts in the other knowledge graph (KG), yet these entities are unlabeled. The problem arises when the source and target graphs are of different scales, and it is much cheaper to label the matchable pairs than the dangling entities. To address this challenge, we propose the framework \textit{Lambda} for dangling detection and entity alignment. Lambda features a GNN-based encoder called KEESA with a spectral contrastive learning loss for EA and a positive-unlabeled learning algorithm called iPULE for dangling detection. Our dangling detection module offers theoretical guarantees of unbiasedness, uniform deviation bounds, and convergence. Experimental results demonstrate that each component contributes to overall performances that are superior to baselines, even when baselines additionally exploit 30\% of dangling entities labeled for training.

Protecting the copyright of large language models (LLMs) has become crucial due to their resource-intensive training and accompanying carefully designed licenses. However, identifying the original base model of an LLM is challenging due to potential parameter alterations. In this study, we introduce HuRef, a human-readable fingerprint for LLMs that uniquely identifies the base model without interfering with training or exposing model parameters to the public. We first observe that the vector direction of LLM parameters remains stable after the model has converged during pretraining, with negligible perturbations through subsequent training steps, including continued pretraining, supervised fine-tuning, and RLHF, which makes it a sufficient condition to identify the base model. The necessity is validated by continuing to train an LLM with an extra term to drive away the model parameters' direction and the model becomes damaged. However, this direction is vulnerable to simple attacks like dimension permutation or matrix rotation, which significantly change it without affecting performance. To address this, leveraging the Transformer structure, we systematically analyze potential attacks and define three invariant terms that identify an LLM's base model. Due to the potential risk of information leakage, we cannot publish invariant terms directly. Instead, we map them to a Gaussian vector using an encoder, then convert it into a natural image using StyleGAN2, and finally publish the image. In our black-box setting, all fingerprinting steps are internally conducted by the LLMs owners. To ensure the published fingerprints are honestly generated, we introduced Zero-Knowledge Proof (ZKP). Experimental results across various LLMs demonstrate the effectiveness of our method. The code is available at https://github.com/LUMIA-Group/HuRef.

With the enrichment of user-item interactions, Graph Neural Networks (GNNs) are widely used in recommender systems to alleviate information overload. Nevertheless, they still suffer from the cold-start issue. Knowledge Graphs (KGs), providing external information, have been extensively applied in GNN-based methods to mitigate this issue. However, current KG-aware recommendation methods suffer from the knowledge imbalance problem caused by incompleteness of existing KGs. This imbalance is reflected by the long-tail phenomenon of item attributes, i.e., unpopular items usually lack more attributes compared to popular items. To tackle this problem, we propose a novel framework called EditKG: Editing Knowledge Graph for Recommendation, to balance attribute distribution of items via editing KGs. EditKG consists of two key designs: Knowledge Generator and Knowledge Deleter. Knowledge Generator generates attributes for items by exploring their mutual information correlations and semantic correlations. Knowledge Deleter removes the task-irrelevant item attributes according to the parameterized task relevance score, while dropping the spurious item attributes through aligning the attribute scores. Extensive experiments on three benchmark datasets demonstrate that EditKG significantly outperforms state-of-the-art methods, and achieves 8.98% average improvement. The implementations are available at https://github.com/gutang-97/2024SIGIR-EditKG.

Large language models (LLMs) have achieved great success in general domains of natural language processing. In this paper, we bring LLMs to the realm of geoscience with the objective of advancing research and applications in this field. To this end, we present the first-ever LLM in geoscience, K2, alongside a suite of resources developed to further promote LLM research within geoscience. For instance, we have curated the first geoscience instruction tuning dataset, GeoSignal, which aims to align LLM responses to geoscience-related user queries. Additionally, we have established the first geoscience benchmark, GeoBench, to evaluate LLMs in the context of geoscience. In this work, we experiment with a complete recipe to adapt a pre-trained general-domain LLM to the geoscience domain. Specifically, we further train the LLaMA-7B model on 5.5B tokens of geoscience text corpus, including over 1 million pieces of geoscience literature, and utilize GeoSignal's supervised data to fine-tune the model. Moreover, we share a protocol that can efficiently gather domain-specific data and construct domain-supervised data, even in situations where manpower is scarce. Meanwhile, we equip K2 with the abilities of using tools to be a naive geoscience aide. Experiments conducted on the GeoBench demonstrate the effectiveness of our approach and datasets on geoscience knowledge understanding and utilization.We open-source all the training data and K2 model checkpoints at https://github.com/davendw49/k2

Generalizing out-of-distribution (OoD) is critical but challenging in real applications such as unmanned aerial vehicle (UAV) flight control. Previous machine learning-based control has shown promise in dealing with complex real-world environments but suffers huge performance degradation facing OoD scenarios, posing risks to the stability and safety of UAVs. In this paper, we found that the introduced random noises during training surprisingly yield theoretically guaranteed performances via a proposed functional optimization framework. More encouragingly, this framework does not involve common Lyapunov assumptions used in this field, making it more widely applicable. With this framework, the upperbound for control error is induced. We also proved that the induced random noises can lead to lower OoD control errors. Based on our theoretical analysis, we further propose OoD-Control to generalize control in unseen environments. Numerical experiments demonstrate the superiority of the proposed algorithm, surpassing previous state-of-the-art by 65% under challenging unseen environments. We further extend to outdoor real-world experiments and found that the control error is reduced by 50% approximately.

Many complex networks in the real world have community structures. Typical examples include online social networks and ecology networks. While the identification of communities bears numerous practical applications, with the increasing awareness of data security and privacy concerns, the need to protect the community affiliations of individuals from disclosing by attackers emerges. This raises the community deception (CD) problem, that is, the opposite of community detection, which asks for ways to minimally perturb the network structures by rewiring nodes so that the target communities maximally hide themself from community detection algorithms. To this end, we investigate the CD problem through a Laplacian spectrum lens and propose a method named$\mathtt {ComDeceptor}$to hide a flexible target set of communities, which is more universal than most existing methods that either focus on hiding the entire communities or a single community. The key idea of$\mathtt {ComDeceptor}$is to first allocate the resources of perturbations fairly and effectively. By proving that hiding communities through intercommunity edge addition and intracommunity edge deletion correspond to maximizing the second smallest eigenvalue$\lambda _{2}$and minimizing the largest eigenvalue$\lambda _{n}$of the graph Laplacian, respectively,$\mathtt {ComDeceptor}$then incorporates efficient heuristics for approximately solving the problems, thus selecting the appropriate edge to perturb. Experimental results over nine real-world networks and six community detection algorithms not only demonstrate the efficiency of$\mathtt {ComDeceptor}$, but also the superior performance on obfuscating community structures over the baselines.

Running on host mobile applications, mini apps have gained increasing popularity these days for its convenience in installation and usage. However, being easy to use allows mini apps to freely access a large amount of user information, mostly without close inspection of privacy violations. Hence it becomes a crucial issue to automatically track sensitive flows in mini apps. Although flow analysis has been widely studied, unique challenges emerge: the analysis tool should not only handle mini app-specific features such as flows that interweave between rendering and logic, and asynchronous executions, but also deal with problems raised by Javascript development: the performance tradeoff between precision and efficiency, and function aliases. To this end, we proposeMiniTracker, an automatic sensitive flow tracking tool which well handles mini app features, constructs assignment flow graphs as common representation across different host apps, searches function aliases, and analyzes the graph by property chains. We show our design choices achieve a sweet spot in the tradeoff between precision and efficiency, with superior performance compared to the state-of-the-art. We also perform a large-scale study on 150 k mini apps, which reveals the common leakage patterns and offers insights into the privacy threats of mini apps.

The robustness of deep neural networks (DNNs) against adversarial example attacks has raised wide attention. For smoothed classifiers, we propose the worst-case adversarial loss over input distributions as a robustness certificate. Compared with previous certificates, our certificate better describes the empirical performance of the smoothed classifiers. By exploiting duality and the smoothness property, we provide an easy-to-compute upper bound as a surrogate for the certificate. We adopt a noisy adversarial learning procedure to minimize the surrogate loss to improve model robustness. We show that our training method provides a theoretically tighter bound over the distributional robust base classifiers. Experiments on a variety of datasets further demonstrate superior robustness performance of our method over the state-of-the-art certified or heuristic methods.

Detecting individualPopulus euphratica (P. euphratica)trees in desert forest areas is crucial for monitoring their ecophysiological characteristics and ecological conservation. However, the presence of the spectral-similarTamarix chinensis (T. chinensis)in the habitats, along with the densely overlapping crowns in clusteredP. euphratica, presents a challenge for the task. This paper proposes a method to detect individualPopulus euphraticain very high spatial resolution (VHR) images. First, the deep learning-based semantic segmentation model is used to differentiate betweenP. euphraticaandT. chinensis. Second, the individual tree detection is converted into a constrained 2D bin packing model and solved by a heuristic template matching and filling algorithm. The experimental data consists of a World View-2 image capturing sparse desert forests of the lower reaches of the Tarim River. 22296 individualP. euphraticatrees were detected, achieving F1 scores of 0.885, 0.869, and 0.902 on three datasets with varying difficulty levels. Furthermore, experiments were conducted to compare with other methods, and the results showed that the proposed method achieved the best performance on all three datasets. The proposed method can be applied to map the distribution of individualP. euphraticatrees in sparse desert forests and can provide methodological references for similar tasks related to individual tree detection in natural forests.

Deep learning technologies have significantly advanced object information extraction from remote sensing data in recent years, achieving broad application across various industrial sectors. However, information loss exists between remote sensing object raster segmentation and geographic information vector mapping, making it challenging to directly apply raster extraction results to vector mapping. This study, taking the automatic extraction of greenhouses based on remote sensing imagery as an example, proposes a vectorization method for remote sensing object segmentation based on frame field. This method bridges the gap between the object pixel segmentation process and the mask vectorization process through the frame field information outputted by the network, resulting in smoother and more regular vector extraction results. To validate the effectiveness of our framework, we introduce the first high-precision greenhouse vector boundary dataset. Extensive experiments demonstrate that our method significantly mitigates the information loss issue prevalent in traditional vectorization processes, achieving a 5.05% improvement in IoU, a 6.06% increase in recall, and a 5.54% reduction in maximum angular error compared to simple vectorization schemes. It outputs more regular greenhouse vector plots, where the precision of the frame field plays a crucial role in the final vectorization quality. This research offers a unique and practical solution, converting remote sensing object segmentation into vector maps.

Graph pooling refers to the operation that maps a set of node representations into a compact form for graph-level representation learning. However, existing graph pooling methods are limited by the power of the Weisfeiler–Lehman (WL) test in the performance of graph discrimination. In addition, these methods often suffer from hard adaptability to hyper-parameters and training instability. To address these issues, we propose Hi-PART, a simple yet effective graph neural network (GNN) framework with Hi erarchical Par tition T ree (HPT). In HPT, each layer is a partition of the graph with different levels of granularities that are going toward a finer grain from top to bottom. Such an exquisite structure allows us to quantify the graph structure information contained in HPT with the aid of structural information theory. Algorithmically, by employing GNNs to summarize node features into the graph feature based on HPT’s hierarchical structure, Hi-PART is able to adequately leverage the graph structure information and provably goes beyond the power of the WL test. Due to the separation of HPT optimization from graph representation learning, Hi-PART involves the height of HPT as the only extra hyper-parameter and enjoys higher training stability. Empirical results on graph classification benchmarks validate the superior expressive power and generalization ability of Hi-PART compared with state-of-the-art graph pooling approaches.

Network alignment, pairing corresponding nodes across the source and target networks, plays an important role in many data mining tasks. Extensive studies focus on learning node embeddings across different networks in a unified space. However, these methods have not taken the large structural discrepancy between aligned nodes into account and, thus, are largely confined by the deterministic representations of nodes. In this work, we propose a novel network alignment framework highlighted by distributional learning and globally optimal alignment. By modeling the uncertainty of each node by Gaussian distribution, our framework builds similarity matrices on the Wasserstein distance between distributions and applies Sinkhorn operation, which learns the globally optimal mapping in an end-to-end fashion. We show that each integrated part of the framework contributes to the overall performance. Under a variety of experimental settings, our alignment framework shows superior accuracy and efficiency to the state-of-the-art.

The great power of Graph Neural Networks (GNNs) relies on a large number of labeled training data, but obtaining the labels can be costly in many cases. Graph Active Learning (GAL) is proposed to reduce such annotation costs, but the existing methods mainly focus on improving labeling efficiency with fixed classes, and are limited to handle the emergence of novel classes. We term the problem as Open-World Graph Active Learning (OWGAL) and propose a framework of the same name. The key is to recognize novel-class as well as informative nodes in a unified framework. Instead of a fully connected neural network classifier, OWGAL employs prototype learning and label propagation to assign high uncertainty scores to the targeted nodes in the representation and topology space, respectively. Weighted sampling further suppresses the impact of unimportant classes by weighing both the node and class importance. Experimental results on four large-scale datasets demonstrate that our framework achieves a substantial improvement of 5.97% to 16.57% on Macro-F1 over state-of-the-art methods.

Graph Neural Network (GNN) has demonstrated extraordinary performance in classifying graph properties. However, due to the selection bias of training and testing data (e.g., training on small graphs and testing on large graphs, or training on dense graphs and testing on sparse graphs), distribution deviation is widespread. More importantly, we often observehybrid structure distribution shiftof both scale and density, despite of one-sided biased data partition. The spurious correlations over hybrid distribution deviation degrade the performance of previous GNN methods and show large instability among different datasets. To alleviate this problem, we proposeOOD-GMixupto jointly manipulate the training distribution withcontrollable data augmentationin metric space. Specifically, we first extract the graph rationales to eliminate the spurious correlations due to irrelevant information. Secondly, we generate virtual samples with perturbation on graph rationale representation domain to obtain potential OOD training samples. Finally, we propose OOD calibration to measure the distribution deviation of virtual samples by leveraging Extreme Value Theory, and further actively control the training distribution by emphasizing the impact of virtual OOD samples. Extensive studies on several real-world datasets on graph classification demonstrate the superiority of our proposed method over state-of-the-art baselines.

Graph Contrastive Learning (GCL) with Graph Neural Networks (GNN) has emerged as a promising method for learning latent node representations in a self-supervised manner. Most of existing GCL methods employ random sampling for graph view augmentation and maximize the agreement of the node representations between the views. However, the random augmentation manner, which is likely to produce very similar graph view samplings, may easily result in incomplete nodal contextual information, thus weakening the discrimination of node representations. To this end, this paper proposes a novel trainable scheme from the perspective of node augmentation, which is theoretically proved to be injective and utilizes the subgraphs consisting of each node with its neighbors to enhance the distinguishability of nodal view. Notably, our proposed scheme tries to enrich node representations via a multi-scale contrastive training that integrates three different levels of training granularity, i.e., subgraph level, graph- and node-level contextual information. In particular, the subgraph-level objective between augmented and original node views is constructed to enhance the discrimination of node representations while graph- and node-level objectives with global and local information from the original graph are developed to improve the generalization ability of representations. Experiment results demonstrate that our framework outperforms existing state-of-the-art baselines and even surpasses several supervised counterparts on four real-world datasets for node classification.

Information cascading, referred as the phenomenon of an individual following the behavior of the preceding individual after observing its actions, is prevalent in real social networks and triggers intense research interests for the purpose of monitoring and controlling network epidemics. One of the typical lines of information cascading study belongs to the Influence Maximization Problem, which aims to algorithmically find the optimal seeds that can spread the information to the maximum number of nodes. Regardless of the tremendous efforts made in various algorithm design of finding such optimal seeds, it has not yet been well understood how the absolute influence power of the “optimal” source set affects the ultimate cascading, i.e., under which conditions the seeds are able or unable to influence an substantial fraction of the entire network. Most existing works have investigated the conditions of network scale influence under linear threshold model, where the activation of a node requires a large number of infected neighbors. Instead, in this paper we focus on the case of single source cascading, which is only possible to occur under the independent cascading model. We launch information cascading analysis from two aspects, i.e., the influence scale and network-scale cascading probability. Firstly, percolation analysis of the cascading outcome shows that estimating influence scale is equivalent to solving fixed point equations. Then, we investigate the speed and stability of information cascading based on fixed point analysis, which shows that the information cascading process almost surely terminates within logarithmic time complexity. Furthermore, the results are generalized to the stochastic block model, where we find that network-scale cascading is determined by the spectral radius of the community matrix. The analysis presented in this paper could help us better understand the conditions for different information cascading outcomes.

This paper presents the design, implementation, and evaluation of TaGroup, a fast, fine-grained, and robust grouping technique for RFIDs. It can achieve a nearly 100% accuracy in distinguishing multiple groups of closely located RFIDs, within only a few seconds. It would benefit many inventory tracking applications, such as self-checkout in retails and packaging quality control in logistics. We make two technical innovations. First, we propose a novel method which can measure the channels between multiple pairs of commercial RFID tags simultaneously, and then estimate the proximity relations between them based on the channel information. Second, we introduce a spatio-temporal graph model which captures a full picture of proximity relations among all the tags, based on which TaGroup can perform a robust grouping of the tags. These two designs together boost the grouping speed and accuracy of TaGroup. Our experiments show that in grouping 120 tags into 4 closely located groups, TaGroup can achieve a nearly 100% accuracy, at the cost of only 2 seconds.

This paper studies connectivity maintenance in uncertain networks under adversarial attack, where a defender conceals crucial links to prevent the largest connected component from being decomposed by an attacker. In contrast with its static counterpart, connectivity maintenance in uncertain networks involves additional probing on links to determine their existence. Therefore, by modeling an uncertain network as a random graph with each link associated with an existence probability and a probing cost, our goal is to design a defensive strategy for link selection that maximizes the expected size of the largest remaining connected component with the minimum expected probing cost, and moreover, the strategy should be independent of the attacking patterns. To this end, we first unravel the computational complexity of the problem by proving its NP-hardness, and then propose optimal defensive strategies based on dynamic programming and multi-objective optimization. Due to the prohibitive computational cost of optimality, two approximate defensive strategies are further designed to pursue decent performance with quasilinear complexity, in which the first one is a heuristic approach that quantifies the link vulnerability through an analogy from the degree centrality of a vertex in static networks to the connectivity weight of a link in uncertain networks, and the second one is an adaptive greedy policy incorporating the minimax rule from game theory, which minimizes the possible loss suffered by the defender in a worst-case scenario and has a constant approximation ratio. Extensive experiments on both synthetic and real-world network datasets under diverse attacking patterns demonstrate the superiority of the proposed strategies over baselines.

This paper presents dual-band Wi-Fi backscatter (DBscatter), which is the first system supporting 2.4GHz and 5GHz Wi-Fi backscatter simultaneously in a single tag. Our key insight is that most existing Wi-Fi devices communicate in the clean 5GHz band. The 5GHz band provides more chances for ”free riding” with less interference, while the 2.4GHz band presents better NLoS performance. DBscatter combines the strengths of the existing 2.4GHz band with the unexplored 5GHz backscatter in a unified design, developing a robust and high-throughput ambient Wi-Fi backscatter system. We make the following technical contributions: (1) We design a dual-band RF frontend to support dual-band Wi-Fi signals. (2) We propose a tag data demodulation algorithm, which merges the common phase error in multi-antenna received signals, improving the tag transmission reliability while reducing the number of required receivers. (3) We build a prototype of DBscatter system using COTS FPGAs and SDRs. Compared to TiScatter and FreeRider, DBscatter boosts Wi-Fi backscatter throughput by 3.74X and 7.35X, and energy efficiency by 1.78X and 1.38X respectively.

We in this paper present TiSee, an RFID-based sensing system that supports miniature robots to perform agile tasks in everyday environments. TiSee’s unique capability is that it uses a single arbitrarily-deployed antenna to locate a target with sub-cm-level accuracy and identify its orientation to within few degrees. Compared with existing solutions which rely on either antenna arrays or multiple RFID readers, TiSee is cheap, compact, and applicable to miniature robots. The idea of TiSee is to stick an RFID tag on the robot (or its gripper) and use it as a moving “antenna” to locate the tags on the target. The core of this design is a novel technique which can build a “channel” between two commercial RFID tags. Such an inter-tag channel is proved to be highly sensitive to the change in inter-tag distance and is resistant to multipath. By leveraging this channel and the mobility of the robot, we emulate an antenna array and use it for fine-grained localization and orientation estimation. Our experiments show that TiSee achieves a median accuracy of 9.5mm and 3.1° in 3D localization and orientation estimation. TiSee brings an eye-in-hand “camera” to miniature robots, supporting them to perform agile tasks in dark, cluttered, and occluded settings.

This paper for the first time demonstrates how to enable OFDMA in Wi-Fi backscatter for capacity and concurrency enhancement. With our design, the excitation signal is reflected, modulated and shifted to lie in the frequency band of the OFDM subcarrier by the tag; OFDMA is realized by coordinating tags to convey information to the receiver with orthogonal subcarriers concurrently through backscatter. The crux of the design is to achieve strict synchronization among communication components, which is more challenging than in regular OFDMA systems due to the more prominent hardware diversity and uncertainty for backscattering. We reveal how the subtle synchronization scenarios particularly for backscattering can incur system offsets, and present a series of novel designs for the excitation signal transmitter, tag, and receiver to address the issue. We build a prototype in 802.11g OFDM framework to validate our design. Experimental results show that our system can achieve 5.2-$16Mbps$aggregate throughput by allowing 48 tags to transmit concurrently, which is 1.45-$5\times $capacity and$48\times $concurrency compared with the existing design respectively. We also design an OFDMA tag IC, with the corresponding simulation and numerical analysis results show that the tag’s power consumption is in tens of$\mu W$.

This article studies time-sensitive multicast in wireless sensor networks (WSNs) with link uncertainty, where information from the source needs to be delivered to multiple receivers within an imposed delay constraint. Prior art on static WSNs minimizes the multicast delay via the construction of a multicast tree that approximates the Steiner tree in length, which, however, may be invalidated by the time-varying network topology of WSNs with uncertain link states. Moreover, for multicast in WSNs with link uncertainty, the possible link failure necessitates a suitable measurement of the uncertain communication distance and calls for the performance guarantee in both delay and delivery ratio. In this work, by modeling a WSN as a random graph with each link associated with a transmission probability, we propose FlowerCast, an efficient multicast scheme, to jointly minimize the expected multicast delay and to maximize the expected delivery ratio of multicast under delay constraint. The core of FlowerCast is to quantify the uncertain communication distance by the expected transmission delay of a time-varying path, based on which a delay-optimal multicast tree is constructed in accordance with the directionality of delay. Candidate paths with a high expected delivery ratio and low expected delay are then selected in a distributed manner to conditionally connect adjacent multicast members and thus transform the multicast tree into a multicast flower. Despite the NP-hardness of optimal candidate paths’ addition, the transformation with the highest expected delivery ratio of multicast under delay constraint can be guaranteed through a pseudo-polynomial time derandomization-based greedy approach. We further demonstrate the time and energy efficiency of FlowerCast through asymptotic analysis. To make full use of the possible overlapping links in a multicast flower, a hybrid routing strategy is presented to wisely switch between sequential routing and synchronous routing for extra enhancement of the multicast performance. Extensive experiments on various datasets verify the superiority of FlowerCast and hybrid routing over baselines and indicate their wide applicability to practical scenarios.

In recent years, multi-agent reinforcement learning (MARL) has presented impressive performance in various applications. However, physical limitations, budget restrictions, and many other factors usually impose constraints on a multi-agent system (MAS), which cannot be handled by traditional MARL frameworks. Specifically, this paper focuses on constrained MASes where agents work cooperatively to maximize the expected team-average return under various constraints on expected team-average costs, and develops a constrained cooperative MARL framework, named DeCOM, for such MASes. In particular, DeCOM decomposes the policy of each agent into two modules, which empowers information sharing among agents to achieve better cooperation. In addition, with such modularization, the training algorithm of DeCOM separates the original constrained optimization into an unconstrained optimization on reward and a constraints satisfaction problem on costs. DeCOM then iteratively solves these problems in a computationally efficient manner, which makes DeCOM highly scalable. We also provide theoretical guarantees on the convergence of DeCOM's policy update algorithm. Finally, we conduct extensive experiments to show the effectiveness of DeCOM with various types of costs in both moderate-scale and large-scale (with 500 agents) environments that originate from real-world applications.

Recent advances in large pre-trained models showed promising results in few-shot learning. However, their generalization ability on two-dimensional Out-of-Distribution (OoD) data, i.e., correlation shift and diversity shift, has not been thoroughly investigated. Researches have shown that even with a significant amount of training data, few methods can achieve better performance than the standard empirical risk minimization method (ERM) in OoD generalization. This few-shot OoD generalization dilemma emerges as a challenging direction in deep neural network generalization research, where the performance suffers from overfitting on few-shot examples and OoD generalization errors. In this paper, leveraging a broader supervision source, we explore a novel Bayesian cross-modal image-text alignment learning method (Bayes-CAL) to address this issue. Specifically, the model is designed as only text representations are fine-tuned via a Bayesian modelling approach with gradient orthogonalization loss and invariant risk minimization (IRM) loss. The Bayesian approach is essentially introduced to avoid overfitting the base classes observed during training and improve generalization to broader unseen classes. The dedicated loss is introduced to achieve better image-text alignment by disentangling the causal and non-casual parts of image features. Numerical experiments demonstrate that Bayes-CAL achieved state-of-the-art OoD generalization performances on two-dimensional distribution shifts. Moreover, compared with CLIP-like models, Bayes-CAL yields more stable generalization performances on unseen classes. Our code is available at https://github.com/LinLLLL/BayesCAL.

Yi Xu, Shuqian Sheng, Jiexing Qi, Luoyi Fu, Zhouhan Lin, Xinbing Wang, Chenghu Zhou. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Researchers usually come up with new ideas only after thoroughly comprehending vast quantities of literature.The difficulty of this procedure is exacerbated by the fact that the number of academic publications is growing exponentially.In this study, we devise a framework based on concept co-occurrence for academic idea inspiration, which has been integrated into a research assistant system.From our perspective, the fusion of two concepts that co-occur in an academic paper can be regarded as an important way of the emergence of a new idea.We construct evolving concept graphs according to the co-occurrence relationship of concepts from 20 disciplines or topics.Then we design a temporal link prediction method based on masked language model to explore potential connections between different concepts.To verbalize the newly discovered connections, we also utilize the pretrained language model to generate a description of an idea based on a new data structure called co-occurrence citation quintuple.We evaluate our proposed system using both automatic metrics and human assessment.The results demonstrate that our system has broad prospects and can assist researchers in expediting the process of discovering new ideas.1 * Luoyi Fu is the corresponding author. 1 The project is publicly available for research purpose https://github.com/xyjigsaw/Kiscovery.

Tianhang Zhang, Lin Qiu, Qipeng Guo, Cheng Deng, Yue Zhang, Zheng Zhang, Chenghu Zhou, Xinbing Wang, Luoyi Fu. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Real-world data usually exhibits a long-tailed distribution, with a few frequent labels and a lot of few-shot labels. The study of institution name normalization is a perfect application case showing this phenomenon: there are many institutions worldwide, with enormous variations of their names in the publicly available literature. In this work, we first collect a large-scale institution name normalization dataset containing over 25k classes whose frequencies are naturally long-tail distributed. We construct our test set from four different subsets: many-, medium-, and few-shot sets, as well as a zero-shot open set, which are meant to isolate the few-shot and zero-shot learning scenarios from the massive many-shot classes. We also replicate several important benchmarks on our data, covering a wide range from search-based methods to neural network methods. Further, we propose our specially pretrained, BERT-based model that shows better out-of-distribution generalization on few-shot and zero-shot test sets. Compared to other datasets focusing on the long-tailed phenomenon, our dataset has one order of magnitude more training data than the largest existing long-tailed datasets and is naturally long-tailed rather than manually synthesized. We believe it provides an important and different scenario to study this problem. To our best knowledge, this is the first natural language dataset that focuses on this long-tailed and open-set classification problem.1

We study the online restless bandit problem, where each arm evolves according to a Markov chain independently, and the reward of pulling an arm depends on both the current state of the corresponding Markov chain and the pulled arm. The agent (decision maker) does not know the transition functions and reward functions, and cannot observe the states of arms even after pulling. The goal is to sequentially choose which arms to pull so as to maximize the expected cumulative rewards collected. In this paper, we propose TSEETC, a learning algorithm based on Thompson Sampling with Episodic Explore-Then-Commit. The algorithm proceeds in episodes of increasing length and each episode is divided into exploration and exploitation phases. During the exploration phase, samples of action-reward pairs are collected in a round-robin fashion and utilized to update the posterior distribution as a mixture of Dirichlet distributions. At the beginning of the exploitation phase, TSEETC generates a sample from the posterior distribution as true parameters. It then follows the optimal policy for the sampled model for the rest of the episode. We establish the Bayesian regret bound $\tilde {\mathcal{O}}(\sqrt{T})$ for TSEETC, where $T$ is the time horizon. We show through simulations that TSEETC outperforms existing algorithms in regret.

User review data is considered as auxiliary information to alleviate the data sparsity problem and improve the quality of learned user/item or interaction representations in review-based recommender systems. However, existing methods usually model user-item interactions in a holistic manner and neglect the entanglement of the latent intents behind them, e.g., price, quality, or appearance, resulting in suboptimal representations and reducing interpretability. In this paper, we propose a Self-supervised Graph Disentangled Networks for review-based recommendation (SGDN), to separately model the user-item interactions based on the latent factors through the textual review data. To this end, we first model the distributions of interactions over latent factors from both semantic information in review data and structural information in user-item graph data, forming several factor graphs. Then a factorized message passing mechanism is designed to learn disentangled user/item and interaction representations on the factor graphs. Finally, we set an intent-aware contrastive learning task to alleviate the sparsity issue and encourage disentanglement through dynamically identifying positive and negative samples based on the learned intent distributions. Empirical results over five benchmark datasets validate the superiority of SGDN over the state-of-the-art methods and the interpretability of learned intent factors.

Data with missing values is ubiquitous in many applications. Recent years have witnessed increasing attention on prediction with only incomplete data consisting of observed features and a mask that indicates the missing pattern. Existing methods assume that the training and testing distributions are the same, which may be violated in real-world scenarios. In this paper, we consider prediction with incomplete data in the presence of distribution shift. We focus on the case where the underlying joint distribution of complete features and label is invariant, but the missing pattern, i.e., mask distribution may shift agnostically between training and testing. To achieve generalization, we leverage the observation that for each mask, there is an invariant optimal predictor. To avoid the exponential explosion when learning them separately, we approximate the optimal predictors jointly using a double parameterization technique. This has the undesirable side effect of allowing the learned predictors to rely on the intra-mask correlation and that between features and mask. We perform decorrelation to minimize this effect. Combining the techniques above, we propose a novel prediction method called StableMiss. Extensive experiments on both synthetic and real-world datasets show that StableMiss is robust and outperforms state-of-the-art methods under agnostic mask distribution shift.

With the popularity of machine learning, it has been a growing concern on the trained model revealing the private information of the training data. Membership inference attack (MIA) poses one of the threats by inferring whether a given sample participates in the training of the target model. Although MIA has been widely studied for discriminative models, for generative models, neither it nor its defense is extensively investigated. In this work, we propose a mixup training method for generative adversarial networks (GANs) as a defense against MIAs. Specifically, the original training data is replaced with their interpolations so that GANs would never overfit the original data. The intriguing part is an analysis from the hypothesis test perspective to theoretically prove our method could mitigate the AUC of the strongest likelihood ratio attack. Experimental results support that mixup training successfully defends the state-of-the-art MIAs for generative models, yet without model performance degradation or any additional training efforts, showing great promise to be deployed in practice.

Overlapping gradient communication with backward computation is a popular technique to reduce communication cost in the widely adopted data parallel S-SGD training. However, the resource contention between computation and All-Reduce communication in GPU-based training reduces the benefits of overlap. With GPU cluster network evolving from low bandwidth TCP to high speed networks, more GPU resources are required to efficiently utilize the bandwidth, making the contention more noticeable. Existing communication libraries fail to account for such contention when allocating GPU threads and have suboptimal performance. In this paper, we propose to mitigate the contention by balancing the overlapped computation and communication time. We formulate an optimization problem that decides the communication thread allocation to reduce overall backward time. We develop a dynamic programming based near-optimal solution and extend it to co-optimize thread allocation with tensor fusion. We conduct simulated study and real-world experiment using an 8-node GPU cluster with 50Gb RDMA network training four representative DNN models. Results show that our method reduces backward time by 10%-20% compared with Horovod-NCCL, by 6%-13% compared with tensor-fusion-optimization-only methods. Simulation shows that our method achieves the best scalability with a training speedup of 1.2x over the best-performing baseline as we scale up cluster size.

Given an aftermath of an information spreading, i.e., an infected network GNafter the propagation of malicious rumors, malware or viruses, how can we identify the sources of the cascade? Answering this problem, which is known as the multiple source detection (MSD) problem, is critical whether for forensic use or insights to prevent future epidemics.Despite the recent considerable effort, most of them are built on a preset propagation model, which limits their application range. Some attempts aim to break this limitation via a label propagation scheme where the nodes surrounded by a large proportion of infected nodes are highlighted. Nonetheless, the detection accuracy may suffer since the node labels are simply integers with all infected or uninfected nodes sharing the same initialization setting respectively, which fall short of sufficiently distinguishing their structural properties. To this end, we propose a community based label propagation (CLP) framework that locates multiple sources through exploiting the community structures formed by infected subgraphs of different sources. Besides, CLP tries to enhance the detection accuracy by incorporating node prominence and exoneration effects, namely the nodes surrounded by larger proportions of infected nodes are more likely to be sources, and the uninfected nodes or infected ones in neighboring community hold the key in exonerating an infected node from being the source. As such, CLP is applicable in more propagation models, and is provably convergent. Experiments on both synthetic and real-world networks further validate the superiority of CLP to the state-of-the-art, boosting the F-score from few percents to approximately 68× in large-scale networks.

Due to the universality of graph data, node classification shows its great importance in a wide range of real-world applications. Despite the successes of Graph Neural Networks (GNNs), GNN based methods rely heavily on rich connections and perform poorly on low-degree nodes. Since many real-world graphs follow a long-tailed distribution in node degrees, they suffer from a substantial performance bottleneck as a significant fraction of nodes is of low degree. In this paper, we point out that under-represented self-representations and low neighborhood homophily ratio of low-degree nodes are two main culprits. Based on that, we propose a novel method Grace which improves the node representation by self-distillation, and increases neighborhood homophily ratio of low-degree nodes by graph completion. To avoid error propagation of graph completion, label propagation is further leveraged. Experimental evidence has shown that our method well supports real-world graphs, and is superior in balancing degree-related bias and overall performance on node classification tasks.

This paper tries to answer a question: "Can we achieve spatial-selective transmission on IoT devices?" A positive answer would enable more secure data transmission among IoT devices. The challenge, however, is how to manipulate signal propagation without relying on beamforming antenna arrays which are usually unavailable on low-end IoT devices.

This paper presents the design, implementation, and evaluation of TaGroup, a fast, fine-grained, and robust grouping technique for RFIDs. It can achieve a nearly 100% accuracy in distinguishing multiple groups of closely located RFIDs, within only a few seconds. It would benefit many inventory tracking applications, such as self-checkout in retails and packaging quality control in logistics.

Reconfigurable reflective arrays can be used to program the radio propagation environment in order to form favorable wireless channel conditions. Previous designs have used large-scale arrays containing hundreds to thousands of reflecting elements located external to the receiving node, with the reflection coefficients of all array elements managed by a controller. However, these designs can be costly to deploy and are challenging to quickly adapt to the time-varying nature of wireless channels caused by mobility.

Road surface markings, like symbols and line markings, are vital traffic infrastructures for driving safety and efficiency. However, real-world conditions can impair the utility of existing road markings. For example, adverse weather conditions such as snow and rain can quickly obliterate visibility.

Bundle recommendation aims to recommend a bundle of items to users as a whole with user-bundle (U-B) interaction information, and auxiliary user-item (U-I) interaction and bundle-item affiliation information. Recent methods usually use two graph neural networks (GNNs) to model user's bundle preferences separately from the U-B graph (bundle view) and U-I graph (item view). However, by conducting statistical analysis, we find that the auxiliary U-I information is far underexplored due to the following reasons: 1) Loosely combining the predicted results cannot well synthesize the knowledge from both views. 2) The local U-B and U-I collaborative relations might not be consistent, leading to GNN's inaccurate modeling of user's bundle preference from the U-I graph. 3) The U-I interactions are usually modeled equally while the significant ones corresponding to user's bundle preference are less emphasized.

Sampling design can significantly reduce the uncertainty in geospatial predictions. In this paper, we developed an adaptive uncertainty-guided stepwise sampling (AUGSS) method to select sampling locations to supplement existing legacy sample points whose representation should be improved. The proposed method selects supplemental samples in a stepwise manner as guided by an objective function with two weighted sub-objectives. One reduces the area with high prediction uncertainty, and the other minimizes the overall prediction uncertainty for the entire area. The method takes an adaptive approach to adjust weights for the two sub-objectives and to tune an uncertainty threshold controlling whether a location can be reliably predicted during the sampling procedure. A case study on soil property prediction shows that AUGSS outperforms the stratified random sampling (SRS) and the non-adaptive uncertainty guided sampling method (UGSS) in terms of RMSE and Lin’s concordance correlation coefficient with different sample sizes. This study shows that the AUGSS method offers a potential for effectively adding supplemental samples to existing samples which are insufficient for spatial prediction. The adaptive strategy guided by predicted uncertainty provides an efficient support to improve the spatial pattern of samples, which plays a key role in the result accuracy of geospatial predictive mapping.

Thin clouds detection and the difficulty in distinguishing between clouds and bright surface features have consistently presented challenges in optical remote sensing cloud detection tasks. Convolutional neural networks (CNNs) have made significant progress, however, CNNs perform weakly in capturing global information interactions due to the inherent limitation of network structure. To address these issues, we propose a hybrid CNN-Transformer network with differential feature enhancement (DFE) for cloud detection (CNN-TransNet). CNN-TransNet adopts a dual-branch encoder consisting of CNN-Transformer module and DFE module. CNN-TransNet combines the strengths of both Transformer and CNN to enhance finer details and build long-range dependencies. CNN is considered as a high-resolution feature extractor for capturing low-level features. The transformer module encodes image sequences by patch embedding to extract high-level features and relationships. DFE branch utilizes differential features and attention mechanism to further obtain effective information for distinguishing between clouds and non-clouds. The decoder upsamples features of the encoder and concatenates multiscale features from the CNN layers. Experimental results demonstrate that the proposed method achieves excellent performance on Landsat-8 and Sentinel-2 images, with a high cloud pixel precision of 92.94% and 93.04%. Moreover, it effectively reduces thin cloud omissions and the misclassifications of bright surface features.

L-band passive microwave remote sensing of ocean surfaces is hampered by uncertainties due to the contribution of precipitation. However, modeling and correcting rain effect are complicated because all the precipitation contributions from the atmosphere and sea surface, including the rain effect in the atmosphere, water refreshing, rain-perturbed sea surface, and rain-induced local wind, are coupled. These rain effects significantly alter L-band microwave satellite measurements. This study investigates the impact of precipitation on the satellite measured brightness temperature (TB) and proposes a correction method. The results show that the rain-induced TB increase is approximately 0.5–1.4 K in the atmosphere, depending on the incidence angle and polarization. Moreover, the rain effect on sea surface emissions is more significant than that in the atmosphere. The result shows that rain effects on sea surface emissions are higher than 3 K for both polarizations when the rain rate is higher than 20 mm/h. We validate the rain effect correction model based on Soil Moisture Active Passive (SMAP) observations. The results show that the TBs at the top of the atmosphere (TOA) simulated by the model are in a good agreement with the SMAP observations, with root-mean-square errors (RMSEs) of 1.137 and 1.519 K for the horizontal and vertical polarizations, respectively, indicating a relatively high accuracy of the established model. Then, a correction model is applied to sea surface salinity (SSS) retrieval for analysis, and the results show that the developed model corrects the underestimation in SSS retrieval. Finally, the rain effect correction model is validated with Aquarius observations in three regions, and it is found that the RMSEs of the corrected TOA TBs range from 1.013 to 1.608 K, which is higher than those without rain effect correction (RMSEs range from 2.078 to 3.894 K). Overall, the model developed in this study provides relatively a good accuracy for rain effect correction.

Verifying the correctness of computation without revealing the input is a critical issue intensively studied in real-world applications. The recent surge of zero knowledge arguments has been focusing on its efficiency and practicality. Among them, GKR-based arguments have received wide attention and become the foundation of many zero-knowledge proof protocols. However, GKR-based protocols are restricted to layered arithmetic circuits. We proposeTerrace, a new, efficient zero-knowledge argument system for general circuits, based on GKR. By dynamically patching cross-layer claims to the original circuit for verification instead of verifying those claims separately,Terraceis able to reduce the total circuit size and thus enjoys a logarithmic factor less verification time and proof size.Terraceis further extended to include the verification of non-arithmetic operations by rewriting those claims in the multilinear extension form. Experimental results demonstrate that Terrace enjoys a competitive performance on efficiency, and shows great promise in enabling low-cost verification of neural networks.

Social network de-anonymization, which refers to re-identifying users by mapping an anonymized network to a correlated real-name network, is an important problem in network science that has received intensive study. Although different de-anonymization algorithms have been proposed, the performance limit of any algorithm on a de-anonymization problem remains less explored. In this paper, we investigate the Optimal De-Anonymization Yield (ODAY), i.e., the maximum extent to which the network can be de-anonymized, of a given de-anonymization problem. Specifically, due to the uncertainty of the true mapping, we define ODAY as the maximum expected number of correctly mapped nodes, and propose general approaches for its quantification in general networks. Therefore, ODAY can be viewed as a quantitative and non-asymptotic concretization of the concept of network de-anonymizability, which is mostly investigated in an asymptotic manner in prior art. Inspired by the effect of network symmetry on the de-anonymizability, we show that for a graph pair with arbitrary topologies, ODAY equals the maximum diagonal sum of a matching probability matrix generated from graph homomorphisms. Due to the exponential complexity of enumerating all the possible homomorphisms, we further obtain an upper bound of ODAY by counting the orbits of each of the two graphs, which significantly reduces the computational cost. Two case studies apply our general findings on symmetry and ODAY to specific network models, and a sampling-based algorithm framework is proposed for ODAY calculation in practice. Extensive experiments are performed to validate our findings. To the best of our knowledge, this work is the first effort that quantifies the de-anonymizability of general networks in a non-asymptotic manner from the perspective of network symmetry, and thus sheds light on privacy enhancement for social network design.

Recent years have witnessed a vastly increasing popularity of location-based social networks (LBSNs), which facilitates studies on the next Point-of-Interest (POI) recommendation problem. A user’s POI visiting behavior shows the sequential transition correlation with previous successive check-ins and the global spatial-temporal correlation with those check-ins that happened a long time ago at a similar time of day and in geographically close areas. Although previous POI recommendation methods attempted to capture these two correlations, several limitations remain to be solved: (1) RNNs are widely adopted to capture the sequential transition correlation, whereas training an RNN is rather time-consuming given the long input check-in sequence. (2) The pairwise proximities on time of day and geographical area of check-ins are crucial for global spatial-temporal correlation learning, but have not been comprehensively considered by previous methods. To tackle these issues, we propose a novel next POI recommendation framework named STA-TCN. Specifically, instead of RNNs, STA-TCN augments the Temporal Convolutional Network with gated input injection to learn sequential transition correlation. Furthermore, STA-TCN fuses two novel grid-difference and time-sensitivity learning mechanisms with attention network to learn the pairwise spatial-temporal proximities among a user’s check-ins. Extensive experiments are conducted on two large-scale real-world LBSN datasets, and the results show that STA-TCN outperforms the best state-of-the-art baseline with an average improvement of 9.71% and 7.88% on hit rate and normalized discounted cumulative gain, respectively.

Self-supervised graph-level representation learning has recently received considerable attention. Given varied input distributions, jointly learning graphs’ unique and common features is vital to downstream tasks. Inspired by graph contrastive learning (GCL), which targets maximizing the agreement between graph representations from different views, we propose an Ada ptive self-supervised framework, Ada-MIP, considering both M utual I nformation between views (unique features) and inter-graph P roximity (common features). Specifically, Ada-MIP learns graphs’ unique information through a learnable and probably injective augmenter, which can acquire more adaptive views compared to the augmentation strategies applied by existing GCL methods; to learn graphs’ common information, we employ graph kernels to calculate graphs’ proximity and learn graph representations among which the precomputed proximity is preserved. By sharing a global encoder, graphs’ unique and common information can be well integrated into the graph representations learned by Ada-MIP. Ada-MIP is also extendable to semi-supervised scenarios, with our experiments confirming its superior performance in both unsupervised and semi-supervised tasks.

Influence maximization through social networks has aroused tremendous interests nowadays. However, people’s various expressions or feelings about a same idea often cause ambiguity via word of mouth. Consequently, the problem of how to maximize the spread of “effective information” still remains largely open. In this paper, we consider a practical setting where ideas can deviate from their original version to invalid forms during message passing, and make the first attempt to seek a union of users that maximizes the spread of effective influence, which is formulated as an Influence Maximization with Information Variation (IMIV) problem. To this end, we model the information as a vector, and quantify the difference of two arbitrary vectors as a distance by a matching function. We further establish a process where such distance increases with the propagation and ensure the recipient whose vector distance is less than a threshold can be effectively influenced. Due to the NP-hardness of IMIV, we greedily select users that can approximately maximize the estimation of effective propagation. Especially, for networks of small scales, we derive a condition under which all the users can be effectively influenced. Our models and theoretical findings are further consolidated through extensive experiments on real-world datasets.

With the increasing popularity of mobile devices, each user is able to conveniently acquire messages from others, and share diverse forms of information, like texts, images, or videos through online mobile apps. The full freedom of speech makes a great amount of truth (i.e., true information) and rumor (i.e., false information) propagate rapidly in a hybrid way through mobile platforms. As a huge variety of information floods pouring over us each day, identifying the authenticity of massive events becomes a necessary task to maintain the stability of Mobile Social Networks (MSNs). An important way to realize it is to trace their diffusions and make judgements according to the reliability of sources. With this regard, this paper proposes a diffusion model that characterizes the simultaneous diffusion of both truth and rumor in realistic MSNs, and makes the first attempt to figure out their respective sources. The problem of interest can be stated as: Given an outcome of cascade of both truth and rumor in MSNs, i.e., a set of nodes that might be the ignorant, the spreader of truth or rumor, or simply the silent receiver, how can we infer both truth sources and rumor sources? Different from previous sources detection works considering single type of nodes, the interplay between truth diffusions and rumor diffusions makes the conventional methods not work. To answer this question, we aim to maximize thesimilarity index, i.e., the number of nodes possessing the same states between the resulting network triggered by our estimated sources with the proposed diffusion model and the given observation network. Compared with existing techniques to trace diffusions of truth or rumor, it is much harder to find two kinds of sets at the same time, including truth sources and rumor sources, due to two primary reasons: (i) our biset optimization makes the submodularity techniques fail; (ii) our objective function is proven to be non-bisubmodular. To overcome above limitations, we first convert the objectivesimilarity indexinto a bisubmodular function by virtue of set covering. Based on this, we propose an approximation algorithm called Truth and Rumor Sources Detection (TRSD) algorithm via multiple reverse samplings with a provable$\frac{1}{4(1+\epsilon)^2}$approximation ratio. Further, a novel “time reversal” sources optimization strategy is proposed to converge the number of output sources from TRSD to a steady state. The effectiveness of our models and algorithms are empirical validated in two various datasets, from which we observe an up to 15% ofsimilarity indexgain as well as a narrowed down gap 0.6% to the ground truth.

The popularity of intelligent devices provides straightforward access to the Internet and online social networks. However, the quick and easy data updates from networks also benefit the risk spreading, such as rumor, malware, or computer viruses. To this end, this article studies the problem of source detection, which is to infer the source node out of an aftermath of a cascade, that is, the observed infected graph G N of the network at some time. Prior arts have adopted various statistical quantities such as degree, distance, or infection size to reflect the structural centrality of the source. In this article, we propose a new metric that we call the infected tree entropy (ITE), to utilize richer underlying structural features for source detection. Our idea of ITE is inspired by the conception of structural entropy [ 21 ], which demonstrated that the minimization of average bits to encode the network structures with different partitions is the principle for detecting the natural or true structures in real-world networks. Accordingly, our proposed ITE based estimator for the source tries to minimize the coding of network partitions brought by the infected tree rooted at all the potential sources, thus minimizing the structural deviation between the cascades from the potential sources and the actual infection process included in G N . On polynomially growing geometric trees, with increasing tree heterogeneity, the ITE estimator remarkably yields more reliable detection under only moderate infection sizes, and returns an asymptotically complete detection. In contrast, for regular expanding trees, we still observe guaranteed detection probability of ITE estimator even with an infinite infection size, thanks to the degree regularity property. We also algorithmically realize the ITE based detection that enjoys linear time complexity via a message-passing scheme, and further extend it to general graphs. Extensive experiments on synthetic and real datasets confirm the superiority of ITE to the baselines. For example, ITE returns an accuracy of 85%, ranking the source among the top 10%, far exceeding 55% of the classic algorithm on scale-free networks.

Despite the bandwidth scaling limit of electrical switching and the high cost of building Clos data center networks (DCNs), the adoption of optical DCNs is still limited. There are two reasons. First, existing optical DCN designs usually face high deployment complexity. Second, these designs are not full-optical and the performance benefit over the non-blocking Clos DCN is not clear. After exploring the design tradeoffs of the existing optical DCN designs, we propose TROD (ThresholdRouting basedOpticalDatacenter), a low-complexity optical DCN with superior performance than other optical DCNs. There are two novel designs in TROD that contribute to its success. First, TROD performs robust topology optimization based on the recurring traffic patterns and thus does not need to react to every traffic change, which lowers deployment and management complexity. Second, TROD introduces tVLB (threshold-based Valiant Load Balance), which can avoid network congestion as much as possible even under unexpected traffic bursts. We conduct simulation based on both Facebook’s real DCN traces and our synthesized highly bursty DCN traces. TROD reduces flow completion time (FCT) by about 1.15-2.16$\times$compared to Google’s Jupiter DCN, at least 2$\times$compared to other optical DCN designs, and about 2.4-3.2$\times$compared to expander graph DCN. Compared with the non-blocking Clos, TROD reduces the hop count of the majority packets by one, and could even outperform the non-blocking Clos with proper bandwidth over-provision at the optical layer. Note that TROD can be built with commercially available hardware and does not require host modifications.

Secure Device-to-Device (D2D) communication is important for the Internet-of-Things (IoT) devices. Key agreement between devices is the important first step in building a secure D2D channel. Due to the lack of third-party certification, key agreement for IoT devices has to rely on the untrusted channel between them, which makes the pairing process vulnerable to attacks such as eavesdropping, jamming, and predictable channel attack. We solve this problem with a novel key agreement method named EchoKey, where two nearby devices can generate a symmetric key independently, leveraging the ambient sound signal that can be locally collected by devices. The intuition of this idea is that the propagation delays of the ambient sounds and the echoes will carry fine-grained information about the spatial context of the receiving device, which can be transformed to a special key for that device. So, nearby devices which have similar spatial context will generate similar echo profiles. We implement a prototype of EchoKey and evaluate its performance in resisting different attacks. The results tell that, with EchoKey, the key agreement process is even undetectable by an attacker which is only 50cm away from the pairing devices, and is thus resistant to all attacks mentioned above.

Jiexing Qi, Jingyao Tang, Ziwei He, Xiangpeng Wan, Yu Cheng, Chenghu Zhou, Xinbing Wang, Quanshi Zhang, Zhouhan Lin. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

With the proliferation of the mobile networks and location-based services, huge volume of user trajectories are collected to analyze the similarity among users and further unveil human mobility patterns for downstream tasks, such as point-of-interest recommendation and tourism planning. In recent works, trajectory embedding methods have been studied as efficient ways of trajectory similarity computation and effective inputs for downstream tasks, which embed trajectories into latent vector spaces equipped with the Euclidean distance to approximate the trajectory similarity and capture the characteristics of human mobility patterns. However, we demonstrate that such embedding, though hiding the locations, can leak the sensitive information of the trajectories, combined with auxiliary data. In this work, we propose trajectory embedding attack schemes to analyze the sensitive information leakage of the embedding vectors. In the experiment, we demonstrate that the passing areas, visited ROIs, and exact shapes of the trajectories are vulnerable under attacks on embedding vectors by the adversary with auxiliary information.

Trajectory representation learning studies the problem of embedding trajectories into low-dimensional vectors, while preserving mutual similarity for the convenience of downstream tasks, such as nearest neighbor search, clustering, classification, etc. In this work, we propose the Trajectory Similarity Network Embedding (TSNE) which exploits representation learning on the k-nearest neighbor partial similarity graph to generate trajectory embeddings, that preserve different similarity efficiently. In theory, we prove that TSNE is equivalent to factorizing the similarity graph, while in practice, TSNE achieves better performance. In the experiment, we show that TSNE outperforms the state-of-the-art baselines, including matrix factorization approaches and RNN based models in terms of similarity preserving and dimension reduction.

High-precision cloud detection for remote sensing image is of great significance for applications in agriculture, environment, meteorology and other fields. Cloud detection in bright surface environments and thin clouds identification have always been a challenge in cloud detection research. Aiming at this problem, a cloud detection method for remote sensing images based on difference features and semantic segmentation network is proposed in this paper. Cloudy and cloudless images of the same area are used to obtain difference features, and cloudless images are used as the surface reference. The multi-band cloud detection network based on U-network (U-net) fully learns the difference feature between clouds and the surface, and combines the feature information of the shallow and deep layers to accurately identify the cloud and the surface. Experiments were conducted on the Landsat 8 validation dataset. The results show that the proposed method achieves good performance. It improves the detection accuracy and reduces the misclassification over the bright surface underlying compared with the method based on top of atmosphere reflectance (TOA).

We in this paper present TiSee, an RFID-based sensing system that supports miniature robots to perform agile tasks in everyday environments. TiSee's unique capability is that it uses a single arbitrarily-deployed antenna to locate a target with sub-cm-level accuracy and identify its orientation to within few degrees. Compared with existing solutions which rely on either antenna arrays or multiple RFID readers, TiSee is cheap, compact, and applicable to miniature robots.

OFDMA downlink design allowing parallel processing OFDM subcarriers is adopted by a number of commercial wireless standards such as LTE, 5G, and 802.11ax. However, the widespread adoption of OFDMA downlink on low-end IoT devices is stymied due to the existing digital receiver framework's ≈100mW power consumption, which is mainly incurred by LO+mixer, ADC, and complex digital processing. In this paper, we present an ultra-low-power OFDMA downlink demodulation design, which achieves ≈100 μW receiving power. Our basic idea is to transform the current digital demodulation approach into the analog one based on filtering, which avoids those power-hungry components. We achieve this by proposing a series of novel RF front-end hardware designs: 1) a μW-level two-stage mixing scheme that enables adjustable and precise subcarrier filtering, 2) a quartz crystal-based filter circuit incurring negligible insertion loss, and 3) a passive phase-to-envelope conversion technique enabling low-power non-coherent phase demodulation. We build a prototype to verify the proposed schemes. Experimental and IC simulation results show that: our new design can achieve 130 -- 1500 times power savings depending on the number of subcarriers that need to be processed in parallel, compared with the traditional all-digital design.

Geographical flows reflect the movements, spatial interactions or connections among locations and are generally abstracted as origin-destination (OD) flows. In this context, clustering is a spatial pattern describing a group of flows with adjacent O and D points. For data composed of two types of flows (bivariate-flow data), a bivariate-flow cluster is a cluster comprising two types of flows, at least one of which exhibits a clustering pattern. In a bivariate-flow cluster, varying flow density combinations imply different meanings. For instance, a cluster with high-density travel flows on both weekdays (type A) and weekends (type B) may be associated with entertainment, whereas high-density flows on weekdays and sparse flows on weekends may reveal work-related travel. However, identifying bivariate-flow clusters with different flow density combinations is still an unsolved problem. To this end, we extend a bivariate-point clustering method and propose a density-based clustering method for bivariate flows. The simulation experiments verify model robustness. In a case study, we apply this method to extract clusters of bivariate-flow data comprising Beijing taxi OD flows of different periods, and identify clusters of work-related, entertainment, tourism, or egress and return travels. These results demonstrate the capability of our method in detecting bivariate-flow clusters.

Tree density is a significant parameter for forest. However, it has always been challenging to use remote-sensing techniques to acquire it efficiently and effectively on a large spatial scale. As a matter of fact, tree counting can be regarded as end-to-end density regression, which uses a remote-sensing image as an input and generates a tree density map as an output. Therefore, the tree number can be easily obtained by the summation of density values in an area. To this end, a deep neural network has been constructed to aggregate multiple decoding paths to extract hierarchical features at different encoding stages, in order to merge tree features of multiple scales in remote-sensing images. At the same time, a hybrid loss function has been proposed to effectively guide the network training and enhance the model ability. A tree count dataset consisting of 2400 sample pairs has been constructed for training and validation. When compared with other popular counting networks, it has been found that the proposed network achieved the best results with a relative mean absolute error (rMAE) of 16.72%, a root mean squared error (RMSE) of 77.96, and an$R^{2}$of 0.96, evidencing that this is a promising method for estimating tree numbers on a large spatial scale.

The von Neumann graph entropy is a measure of graph complexity based on the Laplacian spectrum. It has recently found applications in various learning tasks driven by the networked data. However, it is computationally demanding and hard to interpret using simple structural patterns. Due to the close relation between the Laplacian spectrum and the degree sequence, we conjecture that the structural information, defined as the Shannon entropy of the normalized degree sequence, might be a good approximation of the von Neumann graph entropy that is both scalable and interpretable. In this work, we thereby study the difference between the structural information and the von Neumann graph entropy named asentropy gap. Based on the knowledge that the degree sequence is majorized by the Laplacian spectrum, we for the first time prove that the entropy gap is between 0 and$\log _{2} e$in any undirected unweighted graphs. Consequently we certify that the structural information is a good approximation of the von Neumann graph entropy that achieves provable accuracy, scalability, and interpretability simultaneously. This approximation is further applied to two entropy-related tasks: network design and graph similarity measure, where a novel graph similarity measure and the corresponding fast algorithms are proposed. Meanwhile, we show empirically and theoretically that maximizing the von Neumann graph entropy can effectively hide the community structure, and then propose an alternative metric calledspectral polarizationto guide the community obfuscation. Our experimental results on graphs of various scales and types show that the very small entropy gap readily applies to a wide range of simple/weighted graphs. As an approximation of the von Neumann graph entropy, the structural information is the only one that achieves both high efficiency and high accuracy among the prominent methods. It is at least two orders of magnitude faster than SLaQ (Tsitsulinet al., 2020) with comparable accuracy. Our structural information based methods also exhibit superior performance in downstream tasks such as entropy-driven network design, graph comparison, and community obfuscation.

The research of geoscience plays a strong role in helping people gain a better understanding of the Earth. To effectively represent the knowledge (KG) from enormous geoscience research papers, knowledge graphs can be a powerful means. In the face of enormous geoscience research papers, knowledge graphs can be a powerful means to manage the relationships of data and integrate knowledge extracted from them. However, the existing geoscience KGs mainly focus on the external connection between concepts, whereas the potential abundant information contained in the internal multimodal data of the paper is largely overlooked for more fine-grained knowledge mining. To this end, we propose GAKG, a large-scale multimodal academic KG based on 1.12 million papers published in various geoscience-related journals. In addition to the bibliometrics elements, we also extracted the internal illustrations, tables, and text information of the articles, and dig out the knowledge entities of the papers and the era and spatial attributes of the articles, coupling multimodal academic data and features. Specifically, GAKG realizes knowledge entity extraction under our proposed Human-In-the-Loop framework, the novelty of which is to combine the techniques of machine reading and information retrieval with manual annotation of geoscientists in the loop. Considering the fact that literature of geoscience often contains more abundant illustrations and time scale information compared with that of other disciplines, we extract all the geographical information and era from the geoscience papers' text and illustrations, mapping papers to the atlas and chronology. Based on GAKG, we build several knowledge discovery benchmarks for finding geoscience communities and predicting potential links. GAKG and its services have been made publicly available and user-friendly.

Geographical flow (hereafter flow) can be modeled as an orderly connected point pair composed of an origin (O) and a destination (D). Aggregation is the most common form of spatial heterogeneity of flows, which we define as their deviation from complete spatial randomness (CSR), and the aggregation scale is an important indicator for its perception. Nevertheless, quantifying the aggregation scale of flows is still an unsolved problem. In this paper, we propose the L-function for flows as a solution, derive theoretical null models of the K-function and L-function in a flow space. We conduct simulation experiments to validate the L-function and its capability to detect aggregation scales. Finally, we apply the solution in a case study with taxi data in Beijing and identify nine aggregation scales of taxi OD flows, ranging from 170 m to 22.1 km. These scales correspond to three classes: less than 300 m, from 600 m to 700 m and more than 1500 m. The classes are related to the sizes of the urban facilities where the dominant flow clusters occur, indicating that the L-function in flow space can detect the aggregation scale of flows at the building scale, the block scale and the district scale.

Legacy ecoregion maps contain knowledge on relationships between eco-region units and their environmental factors. This study proposes a method to extract knowledge from legacy area-class maps to formulate a set of fuzzy membership functions useful for regionalization. We develop a buffer zone approach to reduce the uncertainty of boundaries between eco-region units on area-class maps. We generate buffer zones with a Euclidean distance perpendicular to the boundaries, then the original eco-region units without buffer zones serve as the basic units to generate the probability density functions (PDF) of environmental variables. Then, we transform the PDFs to fuzzy membership functions for class-zones on the map. We demonstrate the proposed method with a climatic zone map of China. The results showed that the buffer zone approach effectively reduced the uncertainties of boundaries. A buffer distance of 10–15 km was recommended in this study. The climatic zone map generated based on the extracted fuzzy membership functions showed a higher spatial stratification heterogeneity (compared to the original map). Based on the fuzzy membership functions with climate data of 1961–2015, we also prepared an updated climatic zone map. This study demonstrates the prospects of using fuzzy membership functions to delineate area classes for regionalization purpose.

Satellite-derived nighttime light (NTL) data from the Defense Meteorological Satellite Program's Operational Linescan System (DMSP-OLS) and the Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) have been extensively used for monitoring human activities and urbanization processes. Differences of these two datasets in their spatial and radiometric properties make it difficult for a temporally consistent analysis using these two datasets together. In this article, we developed a new approach to integrate these two datasets and generated a temporally consistent NTL dataset from 1992 to 2018. First, we performed the pixel-level spatial resampling of VIIRS data using a kernel density method after preprocessing the raw VIIRS data. Second, we conducted a logarithmic transformation of the aggregated VIIRS data. Third, we proposed a sigmoid function between DMSP and processed VIIRS data to characterize their relationship. Using the proposed method, we generated a series of consistent DMSP NTL data in Southeast Asia from 1992 to 2018 and analyzed the dynamic of resulted NTL at different scales. The evaluations based on profile curves, spatial patterns, scatter correlations, and histograms, of NTLs, indicate that our approach can achieve a good agreement between DMSP and simulated DMSP data in the same year. Our approach offers the potential for generating a time series of global DMSP NTL data from 1992 to present, which can contribute a more continuous and consistent monitoring of human activities and a better understanding of the urbanization process.

The modelling of human mobility and migration patterns has received much attention due to its substantial importance. Despite long-term efforts, we still lack a modelling framework that captures mobility patterns and further obtains a prospective view of movement trends with regards to diverse impacting factors. Here, we propose a proportional odds model of human mobility and migration (POM-HM) that takes a probabilistic approach to model human movements. Our model is based on the migration probability with a log-logistic distribution under the proportional odds assumption. Explanatory variables are introduced into the model by re-parameterizing the probability distribution function. The two resultant functions, namely, the migration strength and cumulative hazard, are used to estimate regional differences among travel fluxes and their tendencies. The performance of the POM-HM in terms of its validity and accuracy is examined and compared with the gravity model and the radiation model. The probability-based modelling framework enables us to investigate regional variations in migrant fluxes consequently further predict potential future patterns. In short, our modelling approach captures the probabilistic nature of human mobility and migration and furthers our understanding of both the spatiotemporal patterns of population movements and the impacts of various driving forces.

Semantic parsing of large-scale 3D point clouds is an important research topic in computer vision and remote sensing fields. Most existing approaches utilize hand-crafted features for each modality independently and combine them in a heuristic manner. They often fail to consider the consistency and complementary information among features adequately, which makes them difficult to capture high-level semantic structures. The features learned by most of the current deep learning methods can obtain high-quality image classification results. However, these methods are hard to be applied to recognize 3D point clouds due to unorganized distribution and various point density of data. In this paper, we propose a 3DCNN-DQN-RNN method which fuses the 3D convolutional neural network (CNN), Deep Q-Network (DQN) and Residual recurrent neural network (RNN)for an efficient semantic parsing of large-scale 3D point clouds. In our method, an eye window under control of the 3D CNN and DQN can localize and segment the points of the object's class efficiently. The 3D CNN and Residual RNN further extract robust and discriminative features of the points in the eye window, and thus greatly enhance the parsing accuracy of large-scale point clouds. Our method provides an automatic process that maps the raw data to the classification results. It also integrates object localization, segmentation and classification into one framework. Experimental results demonstrate that the proposed method outperforms the state-of-the-art point cloud classification methods.

Remote sensing images often suffer low contrast. Although many contrast enhancement methods have been proposed in recent literature, the efficiency and robustness of remote sensing image contrast enhancement is still a challenge. In this letter, a novel self-adaptive histogram compacting transform-based contrast enhancement method for remote sensing images is presented to meet with the requirements of automation, robustness, and efficiency in applications. First, the histogram of an input image is optimized into compact and continuous status with the constraints of the merging cost, the moderate global brightness, and the entropy contribution of gray levels. Then, a local remapping algorithm is proposed to catch more details during the course of gray extending with the linear stretch. Finally, a dual-gamma transform is proposed to enhance the contrast in both bright and black areas. Experimental and comparison results demonstrate that the proposed method yields better results than the state-of-the-art methods and maintains robustness in different cases. It provides an effective approach for remote sensing image automatic contrast enhancement.

In the high-speed urbanization process of China, the urban population has been increasing significantly, leading to a high-density aggregation of population. However, the sharp increase in population density has not produced commensurate improvements in the road networks. On the contrary, the population increase induced a serious evacuation vulnerability, which cities experience during various hazards and catastrophic events. Therefore, research on evacuation vulnerability is important to urban planning. To assess the evacuation vulnerability, the optimal and worst scenarios should be considered because all possible evacuation plans occur between these extremes. However, most previous evacuation vulnerability studies are based on the worst-case scenario, only providing an upper bound of a potential evacuation assessment. To provide a more comprehensive theoretical basis for decision-makers to understand the consequences caused by all possible evacuations, this paper proposes an optimal evacuation vulnerability assessment model that provides the lower bound on potential evacuation difficulties. The model is solved by a stepwise spreading algorithm based on Graph Theory. Subsequently, to evaluate the effectiveness of the model, the study adopts the model to assess the evacuation capability of different road network topologies. A comparison with previous research was performed. The model was demonstrated in an application to the South Luogu Alley of Beijing, China. The significance of this paper is that the combination of our model with previous research may provide a more complete theoretical basis for an evacuation vulnerability assessment.

When only one type of point is distributed in a region, clustered points can be seen as an anomaly. When two different types of points coexist in a region, they overlap at different places with various densities. In such cases, the meaning of a cluster of one type of point may be altered if points of the other type show different densities within the same cluster. If we consider the origins and destinations (OD) of taxicab trips, the clustering of both in the morning may indicate a transportation hub, whereas clustered origins and sparse destinations (a hot spot where taxis are in short supply) could suggest a densely populated residential area. This cannot be identified by previous clustering methods, so it is worthwhile studying a clustering method for two types of points. The concept of two-component clustering is first defined in this paper as a group containing two types of points, at least one of which exhibits clustering. We then propose a density-based method for identifying two-component clusters. The method is divided into four steps. The first estimates the clustering scale of the point data. The second transforms the point data into the 2D density domain, where the x and y axes represent the local density of each type of point around each point, respectively. The third determines the thresholds for extracting the clusters, and the fourth generates two-component clusters using a density-connectivity mechanism. The method is applied to taxicab trip data in Beijing. Three types of two-component clusters are identified: high-density origins and destinations, high-density origins and low-density destinations, and low-density origins and high-density destinations. The clustering results are verified by the spatial relationship between the cluster locations and their land-use types over different periods of the day.

Land-use classification is essential for urban planning. Urban land-use types can be differentiated either by their physical characteristics (such as reflectivity and texture) or social functions. Remote sensing techniques have been recognized as a vital method for urban land-use classification because of their ability to capture the physical characteristics of land use. Although significant progress has been achieved in remote sensing methods designed for urban land-use classification, most techniques focus on physical characteristics, whereas knowledge of social functions is not adequately used. Owing to the wide usage of mobile phones, the activities of residents, which can be retrieved from the mobile phone data, can be determined in order to indicate the social function of land use. This could bring about the opportunity to derive land-use information from mobile phone data. To verify the application of this new data source to urban land-use classification, we first construct a vector of aggregated mobile phone data to characterize land-use types. This vector is composed of two aspects: the normalized hourly call volume and the total call volume. A semi-supervised fuzzy c-means clustering approach is then applied to infer the land-use types. The method is validated using mobile phone data collected in Singapore. Land use is determined with a detection rate of 58.03%. An analysis of the land-use classification results shows that the detection rate decreases as the heterogeneity of land use increases, and increases as the density of cell phone towers increases.

The demand for parallel geocomputation based on raster data is constantly increasing with the increase of the volume of raster data for applications and the complexity of geocomputation processing. The difficulty of parallel programming and the poor portability of parallel programs between different parallel computing platforms greatly limit the development and application of parallel raster-based geocomputation algorithms. A strategy that hides the parallel details from the developer of raster-based geocomputation algorithms provides a promising way towards solving this problem. However, existing parallel raster-based libraries cannot solve the problem of the poor portability of parallel programs. This paper presents such a strategy to overcome the poor portability, along with a set of parallel raster-based geocomputation operators (PaRGO) designed and implemented under this strategy. The developed operators are compatible with three popular types of parallel computing platforms: graphics processing unit supported by compute unified device architecture, Beowulf cluster supported by message passing interface (MPI), and symmetrical multiprocessing cluster supported by MPI and open multiprocessing, which make the details of the parallel programming and the parallel hardware architecture transparent to users. By using PaRGO in a style similar to sequential program coding, geocomputation developers can quickly develop parallel raster-based geocomputation algorithms compatible with three popular parallel computing platforms. Practical applications in implementing two algorithms for digital terrain analysis show the effectiveness of PaRGO.

This research presented a framework to track and query spatiotemporal changes and interactions of dynamic geographic phenomena. The framework organized information of dynamic phenomena as a hierarchy of static structures, processes, and scenarios. Static structures of a dynamic phenomenon at its different evolution stages were described by its corepoint, footprint border, and composite border, which were extracted from time series remote sensing images. Time series static structures of a phenomenon were then grouped into processes to show its changes over space and time. Scenarios were used to describe a collection of interacting processes in space. We expanded the identity-based change (IBC) model by adding more primitives and operations to represent semantics of these changes and interactions. A geographic information system (GIS) database was built by integrating the expanded IBC model with our spatiotemporal framework. As demonstrated by a case study of ocean eddies in the South China Sea (SCS), query results of the behaviors and relationships of ocean eddies from the GIS database help us better understand their development and evolution, demonstrating the usefulness of this spatiotemporal framework. Spatial and semantic queries about a specific eddy from the database can further efficiently present its lifetime dynamic changes and all other eddies that interacted with it.

The estimation of regional gross primary production (GPP) is a crucial issue in carbon cycle studies. One commonly used way to estimate the characteristics of GPP is to infer the total amount of GPP by collecting field samples. In this process, the spatial sampling design will affect the error variance of GPP estimation. This letter uses geostatistical model-based sampling to optimize the sampling locations in a spatial heterogeneous area. The approach is illustrated with a real-world application of designing a sampling strategy for estimating the regional GPP in the Babao river basin, China. By considering the heterogeneities in the spatial distribution of the GPP, the sampling locations were optimized by minimizing the spatially averaged interpolation error variance. To accelerate the optimization process, a spatial simulated annealing search algorithm was employed. Compared with a sampling design without considering stratification and anisotropies, the proposed sampling method reduced the error variance of regional GPP estimation.

Selectivity estimation is crucial to query optimizers in choosing an optimal execution plan in a given spatial query, and there has been a great deal of focus on how to achieve good selectivity estimation for finer spatial selection operators. Equally crucial to this is understanding how to produce an updated spatial histogram. With this in mind, we used a cumulative annular bucket histogram (AB histogram), which not only accurately estimates the selectivity of a spatial selection or a spatial join operation with finer operators but also provides an updated spatial histogram to estimate the selectivity of subsequent spatial operations in a multi-level spatial query plan. A basic unit of AB histogram stores the number of minimum bounding rectangles whose lower left points and upper right points are located in specific rectangular regions. According to the basic units of a cumulative AB histogram, we can find out the selectivity of a spatial selection with a number of different finer operators. When it comes to spatial join operations, a relationship between two cumulative AB histograms can be translated into a relationship between one histogram and numerous query windows from the other histogram. Furthermore, an updated cumulative AB histogram can be simultaneously built into the process of selectivity calculation, making it possible to achieve both selectivity and an updated histogram of spatial join; its implementation made in the optimizer facility (OPF) of INGRES9.2. To highlight the performance of a cumulative AB histogram, several experiments have been conducted, with results showing that the cumulative AB histogram not only supports the selectivity estimation of spatial selection and spatial join with ‘Disjoint’, ‘Intersect’, ‘Within’, ‘Contains’, ‘Crosses’ and ‘Overlap’ operators but also supports the generation of an updated histogram. This indicates that Ingres would do better to find a query plan with low-execution costs.

A temporal point process is a sequence of points, each representing the occurrence time of an event. Each temporal point process is related to the behavior of an entity. As a result, clustering of temporal point processes can help differentiate between entities, thereby revealing patterns of behaviors. This study proposes a hierarchical cluster method for clustering temporal point processes based on the discrete Fréchet (DF) distance. The DF cluster method is divided into four steps: (1) constructing a DF similarity matrix between temporal point processes; (2) constructing a complete linkage hierarchical tree based on the DF similarity matrix; (3) clustering the point processes with a threshold determined by locating the local maxima on the curve of the pseudo-F statistic (an index which measures the separability between clusters and the compactness in clusters); and (4) identifying inner patterns for each cluster formed by a series of dense intervals, each of which contains at least one event of all processes of the cluster. The contributions of the article are: (1) the proposed DF cluster method can cluster temporal point processes into different groups and (2) more importantly, it can identify the inner pattern of each cluster. Two synthetic data sets were created to illustrate the DF distance between temporal point process clusters (the first data set) and validate the proposed DF cluster method (the second data set), respectively. An experiment and a comparison with a method based on dynamic time warping show that DF cluster successfully identifies the preconfigured patterns in the second synthetic data set. The cluster method was then applied to a population migration history data set for the Northern Plains of the United States, revealing some interesting population migration patterns.

In this study, we proposed a new method for extracting urban built-up area from satellite-based nighttime light data. Our methods are based on locally thresholding correction of nighttime lights in association with economic statistical data (termed TC-ESD). We grouped China's cities into a series of categories according to gross domestic product and geographical regions. Category-based thresholds of night light brightness used in extracting urban extent were obtained for different categories of cities from a comprehensive consideration of nighttime lights and economic activities. In comparison with uniform empirical thresholding method, our results show that TC-ESD method could markedly improve the recognition accuracy of urban built-up area from nightlight data for individual cities.

In the map of geo-referenced population and cases, the detection of the most likely cluster (MLC), which is made up of many connected polygons (e.g., the boundaries of census tracts), may face two difficulties. One is the irregularity of the shape of the cluster and the other is the heterogeneity of the cluster. A heterogeneous cluster is referred to as the cluster containing depression links (a polygon is a depression link if it satisfies two conditions: (1) the ratio between the case number and the population in the polygon is below the average ratio of the whole map; (2) the removal of the polygon will disconnect the cluster). Previous studies have successfully solved the problem of detecting arbitrarily shaped clusters not containing depression links. However, for a heterogeneous cluster, existing methods may generate mistakes, for example, missing some parts of the cluster. In this article, a spatial scanning method based on the ant colony optimization (AntScan) is proposed to improve the detection power. If a polygon can be simplified as a node, the research area consisting of many polygons then can be seen as a graph. So the detection of the MLC can be seen as the search of the best subgraph (with the largest likelihood value) in the graph. The comparison between AntScan, GAScan (the spatial scan method based on the genetic optimization), and SAScan (the spatial scan method based on the simulated annealing optimization) indicates that (1) the performance of GAScan and SAScan is significantly influenced by the parameter of the fraction value (the maximum allowed size of the detected cluster), which can only be estimated by multiple trials, while no such parameter is needed in AntScan; (2) AntScan shows superior power over GAScan and SAScan in detecting heterogeneous clusters. The case study on esophageal cancer in North China demonstrates that the cluster identified by AntScan has the larger likelihood value than that detected by SAScan and covers all high-risk regions of esophageal cancer whereas SAScan misses some high-risk regions (the region in the southwest of Shandong province, eastern China) due to the existence of a depression link.

In a spatio-temporal data set, identifying spatio-temporal clusters is difficult because of the coupling of time and space and the interference of noise. Previous methods employ either the window scanning technique or the spatio-temporal distance technique to identify spatio-temporal clusters. Although easily implemented, they suffer from the subjectivity in the choice of parameters for classification. In this article, we use the windowed kth nearest (WKN) distance (the geographic distance between an event and its kth geographical nearest neighbour among those events from which to the event the temporal distances are no larger than the half of a specified time window width [TWW]) to differentiate clusters from noise in spatio-temporal data. The windowed nearest neighbour (WNN) method is composed of four steps. The first is to construct a sequence of TWW factors, with which the WKN distances of events can be computed at different temporal scales. Second, the appropriate values of TWW (i.e. the appropriate temporal scales, at which the number of false positives may reach the lowest value when classifying the events) are indicated by the local maximum values of densities of identified clustered events, which are calculated over varying TWW by using the expectation-maximization algorithm. Third, the thresholds of the WKN distance for classification are then derived with the determined TWW. In the fourth step, clustered events identified at the determined TWW are connected into clusters according to their density connectivity in geographic–temporal space. Results of simulated data and a seismic case study showed that the WNN method is efficient in identifying spatio-temporal clusters. The novelty of WNN is that it can not only identify spatio-temporal clusters with arbitrary shapes and different spatio-temporal densities but also significantly reduce the subjectivity in the classification process.

Most multiple‐flow‐direction algorithms (MFDs) use a flow‐partition coefficient (exponent) to determine the fractions draining to all downslope neighbours. The commonly used MFD often employs a fixed exponent over an entire watershed. The fixed coefficient strategy cannot effectively model the impact of local terrain conditions on the dispersion of local flow. This paper addresses this problem based on the idea that dispersion of local flow varies over space due to the spatial variation of local terrain conditions. Thus, the flow‐partition exponent of an MFD should also vary over space. We present an adaptive approach for determining the flow‐partition exponent based on local topographic attribute which controls local flow partitioning. In our approach, the influence of local terrain on flow partition is modelled by a flow‐partition function which is based on local maximum downslope gradient (we refer to this approach as MFD based on maximum downslope gradient, MFD‐md for short). With this new approach, a steep terrain which induces a convergent flow condition can be modelled using a large value for the flow‐partition exponent. Similarly, a gentle terrain can be modelled using a small value for the flow‐partition exponent. MFD‐md is quantitatively evaluated using four types of mathematical surfaces and their theoretical ‘true’ value of Specific Catchment Area (SCA). The Root Mean Square Error (RMSE) shows that the error of SCA computed by MFD‐md is lower than that of SCA computed by the widely used SFD and MFD algorithms. Application of the new approach using a real DEM of a watershed in Northeast China shows that the flow accumulation computed by MFD‐md is better adapted to terrain conditions based on visual judgement.

When two spatial point processes are overlaid, the one with the higher rate is shown as clustered points, and the other one with the lower rate is often perceived to be background. Usually, we consider the clustered points as feature and the background as noise. Revealing these point clusters allows us to further examine and understand the spatial point process. Two important aspects in discerning spatial cluster features from a set of points are the removal of noise and the determination of the number of spatial clusters. Until now, few methods were able to deal with these two aspects at the same time in an automated way. In this study, we combine the nearest‐neighbour (NN) method and the concept of density‐connected to address these two aspects. First, the removal of noise can be achieved using the NN method; then, the number of clusters can be determined by finding the density‐connected clusters. The complexity for finding density‐connected clusters is reduced in our algorithm. Since the number of clusters depends on the value of k (the kth nearest neighbour), we introduce the concept of lifetime for the number of clusters in order to measure how stable the segmentation results (or number of clusters) are. The number of clusters with the longest lifetime is considered to be the final number of clusters. Finally, a seismic example of the west part of China is used as a case study to examine the validity of our method. In this seismic case study, we discovered three seismic clusters: one as the foreshocks of the Songpan quake (M = 7.2), and the other two as aftershocks related to the Kangding‐Jiulong (M = 6.2) quake and Daguan quake (M = 7.1), respectively. Through this case study, we conclude that the approach we proposed is effective in removing noise and determining the number of feature clusters.

In this paper, the method of triangles slice construction was presented for reconstructing 3D pipeline surface. In this means, shape points were calculated by surface outline equation which was got from quadratic surface. Then these shape points were patched with triangles grid to constructed triangles slice, which gained the pipeline surface model. Finally, the pipeline models were rendered by means of computer graphics in order to implement 3D pipeline surface reconstruction. In the end, a reconstruction instance was given.

In this paper, GML and SVG are discussed firstly, especially in the role of open WebGIS. Then an open WebGIS framework is put forward that using GML to store and integrate geographic spatial data and using SVG to implement Web display and operation for geographic spatial data. In the end, taking a forestry data as an example, an open WebGIS application system was set up and discussed.

Vector-based association rule discovery models have been provided to look for spatial knowledge in terrestrial applications recently. Most of them just consider the attribution relationship of local space or the topological relationship between parcels or objects. And they are difficult to consider the temporal change, distance relationship or direction relationship. In this presentation, an geoevent association rule discovery model (GEARDM) is developed, which is based on the spatiotemporal grid and rough set, and the experiment in marine fishery application is provided in support of the model. In GEARDM, the continuous spatiotemporal process is discretized by fuzzy preknowledge as spatiotemporal assignment and will be assigned into a decision table. The mining algorithm based on rough set results from the decision table in the geoevent association rules, which shows what kind of pattern of spatiotemporal assignment of environmental factors determine the happen or attribution of the geoevents. After the mined rules being analyzed by knowledge, the refined knowledge is obtained and the spatiotemporal assignment will be renewed. Then the flow of GEARDM feeds back to the phase of discretization and iterates until the final rules are reasonable. These knowledge-based rules also can be used in the expert system to predict the happen or the scale of geoevent as the experiments show.

The general image segmentation approach used in other domains may not be applicable to the remote sensing field, which is due to the following factors: remotely sensed data is multi-spectral, always very large in size, and in multi-scale as well. How to quickly and efficiently segment remotely sensed imagery is still a big issue to be solved. Based on human vision mechanism, a new hybrid multi-scale segmentation approach is presented, which is implemented at three coarse-to-fine scale levels. First, remotely sensed imagery is segmented at a coarse scale, and image regions (segments) are produced. Then, the corresponding regions in the original image are segmented by another segmentation approach one by one at the fine scale. From the experiment results, we found the approach is rather promising. However, there still exists some problems to be settled, and further researches should be conducted in the future.

Some results of analysis conclude that the internal wave can affect the sea wind-generated wave crucially, so that the spectrum accounts for the interactions between the two kinds of sea water masses. These backscattering coefficients of the internal wave sea surface of the South China Sea are calculated using some existing models and the results are compared with the calculations of SIR-C/X-SAR data. Different results among the radar bands exist and are affected by sea surface roughness. The ocean wave spectrum is derived from the X-SAR image. Finally, we point out that the tidal current is the main factor that causes the internal waves.