Multi-view graph clustering (MVGC) for remote sensing data has gained increasing attention due to its ability to integrate complementary information across modalities while capturing spatial dependencies in heterogeneous data. Although current methods based on graph contrastive learning achieve strong performance, they often misidentify intra-cluster samples as negatives, leading to class conflicts and reduced clustering accuracy. Graph masked autoencoders have recently shown promising potential in learning robust representations through masked reconstruction, but their application to remote sensing data remains underexplored. This challenge is especially notable in the multi-view remote sensing setting, where high heterogeneity and complex spatial structures increase the difficulty of effective representation learning. To address these issues, we propose Clustering-Guided graph Mask AutoEncoder (CG-MAE), the first framework to extend graph masked autoencoders to multi-view remote sensing clustering. We introduce a clustering-guided masking strategy that selectively masks nodes near cluster centers and intra-cluster edges, which are crucial for capturing key structural information. By reconstructing these masked components, the model is encouraged to focus on learning features that are highly relevant to clustering. To further improve training stability and efficiency, we design an easy-to-hard node masking strategy that enables the model to gradually learn from increasingly challenging patterns. Additionally, we propose a dual self-adaptive learning mechanism that encourages the model to align more closely with the underlying semantic distributions. Extensive experiments on four widely used multi-view remote sensing datasets demonstrate that CG-MAE consistently outperforms state-of-the-art methods in both clustering accuracy and representation quality.

Graph Neural Networks (GNNs) are vulnerable to graph backdoor attacks, which poses severe risks to their deployment in safety-critical applications. Existing defenses predominantly focus on specific backdoor triggers, making them brittle and unable to generalize across different backdoor triggers with varying properties. Motivated by this limitation, this work proposes an attack-agnostic graph backdoor defense mechanism A2GBD, which does not require prior knowledge of the specific attack strategies (e.g., edge perturbation, node attribute manipulation) to achieve effective defense. A2GBD consists of suspicious node selection and defense strategy generation. The selection module selects high-suspicion nodes to enhance defense awareness, while the defense agent adaptively determines and executes defense strategies. Extensive experiments on multiple benchmark datasets demonstrate that A2GBD consistently lowers attack success rates while maintaining high clean accuracy, showing strong robustness and generalizability against diverse graph backdoor attack strategies.

Pairwise learning underpins implicit collaborative filtering, yet its effectiveness is often hindered by sparse supervision, noisy interactions, and popularity-driven exposure bias. In this paper, we propose Variational Bayesian Personalized Ranking (VarBPR), a tractable variational framework for implicit-feedback pairwise learning that offers principled exposure controllability and theoretical interpretability. VarBPR reformulates pairwise learning as variational inference over discrete latent indexing variables, explicitly modeling noise and indexing uncertainty, and divides training into two stages: variational inference, which solve variational posteriors, and variational learning, which updates model parameters based on these posteriors. In the variational inference stage, we develop a variational formulation that integrates preference alignment, denoising, and popularity debiasing under a unified ELBO/regularization objective, deriving closed-form posteriors with clear control semantics: the prior encodes a target exposure pattern, while temperature/regularization strength controls posterior-prior adherence. As a result, exposure controllability becomes an endogenous and interpretable outcome of variational inference. In the variational learning stage, we propose a posterior-compression objective that reduces the ideal ELBO's computational complexity from polynomial to linear, with the approximation justified by an explicit Jensen-gap upper bound. Theoretically, we provide interpretable generalization guarantees by identifying a structural error component and revealing the opportunity cost of prioritizing certain exposure patterns (e.g., long-tail), offering a concrete analytical lens for designing controllable recommender systems. Empirically, We validate VarBPR across popular backbones; it demonstrates consistent gains in ranking accuracy, enables controlled long-tail exposure, and preserves the linear-time complexity of BPR.

Understanding the spatiotemporal characteristics of criminal activities in a city, or urban crime dynamics for short, is essential for developing ways to control crime and improve urban safety. While much effort has been devoted to this field, most of the existing studies have led to overly generalized findings, obscuring the ways in which dynamic patterns of criminal activities vary by place, time, and situational context. To address this challenge, this article proposes a novel space-time-context-aware tensor decomposition framework, namelySTCTD-Crime, for an in-depth understanding of urban crime dynamics. Specifically,STCTD-Crimefirst constructs a third-order tensor to represent crime data, which provides an elegant way to model spatial, temporal, and contextual factors simultaneously. Then, it decouples the influence that the three factors exerts on criminal activities via the tensor decomposition, enabling the observation of the extent to which each factor affects crime incidents occurring at different regions, within different time slices, and under different situational contexts. Moreover,STCTD-Crimeexploits spatiotemporal correlations between criminal activities to facilitate the understanding of dynamics by seamlessly integrating a crime-number-guided correlation learning method into the framework. Finally, an alternating optimization based scheme is developed to solve the optimization problem, which results in an efficient urban crime dynamics discovery procedure. Extensive analyses on crime datasets drawn from real-world sources convincingly demonstrate the effectiveness ofSTCTD-Crime.

Coclustering, as a crucial technique in data analysis, has proven effective in exploring the inherent data structures and revealing the synergistic effects of samples and features. However, most existing coclustering methods are limited to graph structures that primarily capture pairwise relationships between data points. In contrast, hyperedges in hypergraphs can connect multiple nodes, capturing complex associations between sets of points. Compared with the unidirectional clustering of hyperedges and nodes, bidirectional clustering of samples and features on hypergraphs presents a greater challenge. Moreover, unidirectional clustering overlooks the mutual dependencies between rows and columns, making it difficult to optimize clustering performance. Therefore, this article proposes a novel general and adaptive hypergraph model for coclustering (GAHGC), aiming to cocluster samples and features. It not only captures the similarity between samples but also considers the correlation information between samples and features. First, a general hypergraph construction strategy is proposed to overcome the limitation of existing methods that are often constrained by specific hypergraph data types. Next, an adaptive hypergraph optimization mechanism is designed to enhance coclustering performance. Finally, extensive comparative experiments on benchmark datasets demonstrate the effectiveness and superiority of the proposed model, further revealing the potential of hypergraph coclustering in capturing complex multilevel interactions.

Although multi-view unsupervised feature selection is a promising technique for dimensionality reduction on unlabeled multi-view data, existing methods cannot directly address incomplete data, where certain samples are missing in specific views. These methods typically begin by imputing missing data using predetermined values, followed by performing feature selection on the completed dataset. However, the separation of imputation and feature selection processes fails to exploit their inherent synergy, as local structural information obtained from feature selection could guide the imputation process and, in turn, improve the overall effectiveness of feature selection. In addition, previous methods rely on similarity graphs based on Euclidean distance to preserve the local manifold structure but overlook the topological relationships within the data, thereby hindering accurate capture of intrinsic structures. In this paper, we propose an adaptive topological similarity learning for incomplete multi-view unsupervised feature selection method (ATSL-IMUFS) to address the aforementioned issues. ATSL-IMUFS first integrates multi-view feature selection and missing data imputation into a unified learning framework. Then, it adaptively learns similarity graphs for each view while simultaneously capturing the consensus topological relationship across views, effectively characterizing the local manifold structure. Extensive experiments conducted on real-world datasets demonstrate the superior performance of ATSL-IMUFS compared to competing methods.

Hashing-based methods have gained widespread attention in cross-modal retrieval due to their significant potential for spatiotemporal efficiency and query accuracy, driving significant progress. However, most existing algorithms extract similarity information solely from labels, overlooking the consistent semantic similarities embedded in features. Additionally, some algorithms underutilize label semantics, neglecting inherent category relationships and their interplay with feature semantics, which compromises algorithmic effectiveness. Furthermore, current methods often suffer from weak discriminative power and semantic misalignment when embedding similarity information into hash codes, leading to learning biases. To address these short-comings, a Multi-similarity Preserving Semantic-aware Hashing (MPSH) algorithm is proposed. MPSH first learns a shared semantic space from heterogeneous features to generate dynamic feature fusion similarity. Subsequently, it explores label distributions to construct category vectors, which are collaboratively associated with isomorphic feature representations to form a semantic-aware space. Finally, manifold embedding is employed to effectively preserve multi-similarity and data distribution, yielding semantically rich hash representations. Moreover, a non-relaxed discrete optimization algorithm is introduced to accelerate hash code iteration. Extensive experiments on multiple benchmark datasets, comparing MPSH with leading algorithms, demonstrate its superior efficiency and performance.

As a core problem in unsupervised learning, anomaly detection focuses on identifying abnormal patterns in datasets, thereby providing support for uncovering potential problems and extracting valuable information. However, most existing methods fail to extract sufficient information in feature interactions when dealing with heterogeneous datasets. To address this challenge, a novel anomaly detection method based on multi-sequence fuzzy feature interaction is proposed. Firstly, we propose multi-sequence features based on joint fuzzy information entropy to capture complex feature interactions and to quantify the interdependencies among features. Secondly, forward and reverse multi-sequence feature subset pairs are constructed to characterize the correlation between features from different angles, enhancing the accuracy of representing complex interactions in heterogeneous data and improving the ability to identify potential anomalies. Subsequently, an uncertainty measure based on multi-sequence information fusion is introduced, and anomaly scores are accumulated by incorporating instance weights, thereby ensuring stable detection performance in heterogeneous datasets. Finally, an anomaly detection algorithm based on multi-sequence fuzzy feature interaction (ADMSFI) is proposed. The experimental results demonstrate that the proposed algorithm ADMSFI significantly outperforms 13 existing algorithms in terms of performance and flexibility in 24 datasets.

Unsupervised feature selection (UFS) has recently gained attention for its effectiveness in processing unlabeled high-dimensional data. However, existing methods overlook the intrinsic causal mechanisms within the data, resulting in the selection of irrelevant features and poor interpretability. Additionally, previous graph-based methods fail to account for the differing impacts of non-causal and causal features in constructing the similarity graph, which leads to false links in the generated graph. To address these issues, a novel UFS method, called Causally-Aware UnSupErvised Feature Selection learning (CAUSE-FS), is proposed. CAUSE-FS introduces a novel causal regularizer that reweights samples to balance the confounding distribution of each treatment feature. This regularizer is subsequently integrated into a generalized unsupervised spectral regression model to mitigate spurious associations between features and clustering labels, thus achieving causal feature selection. Furthermore, CAUSE-FS employs causality-guided hierarchical clustering to partition features with varying causal contributions into multiple granularities. By integrating similarity graphs learned adaptively at different granularities, CAUSE-FS increases the importance of causal features when constructing the fused similarity graph to capture the reliable local structure of data. Extensive experimental results demonstrate the superiority of CAUSE-FS over state-of-the-art methods, with its interpretability further validated through feature visualization.

Representation learning serves as a critical bridge between human cognition and the data world, constituting an essential component of machine learning architectures where comprehensiveness and flexibility are paramount. However, existing multi-view feature selection methods are constrained by the raw-scale representations, neglecting comprehensive depth-breadth integration and flexible adaptability. This paper presents multi-scale fuzzy fusion-based heterogeneous granular-ball flexible representation learning for multi-view feature selection (MFHGBFR). A comprehensive multi-view, multi-scale analytical foundation is established to promote depth-breadth representation learning, where views represent breadth and scales represent depth in human cognition. Heterogeneous granular ball-based flexible representations are adaptively developed via multi-scale fuzzy fusion, effectively capturing intricate data manifolds and implicit fuzzy patterns across multi-granularity spaces. Intricate data manifolds are fitted with heterogeneous rather than conventional homogeneous structures to improve the adaptability of representations' multi-granularity. Implicit fuzzy patterns are extracted with fuzzy approximation operators to mitigate fuzziness and uncertainty in a multi-granularity space. For the first time, granular-ball representation learning and feature selection are adaptively optimized in a unified one-step framework, rather than the conventional two-step frameworks, for flexible adaptability. An effective optimization algorithm with proven convergence is derived. Through the learned high-quality representations, the MFHGBFR manifests superior performance, robustness, and efficiency. Comparative experiments against contemporary state-of-the-art algorithms substantiate these advantages quantitatively and qualitatively.

Traffic forecasting is pivotal but challenging due to intricate spatio-temporal dynamics. Existing models often apply a uniform spatial mechanism across distinct temporal scales and rely on static feature embeddings. Consequently, they are inadequate in capturing scale-specific spatial heterogeneity and dynamic feature interdependencies. To address these limitations, we propose the Multi-Scale Spatio-Temporal Attention Network (MSSTAN) with a novel dual-branch architecture: (1) A Global-Local Feature Attention Network (GLFAN) that explicitly decouples spatial interactions across decomposed temporal components to capture multi-scale spatial patterns; and (2) A Spatio-Temporal Feature Attention Network (STFAN) that dynamically recalibrates feature importance based on specific spatio-temporal contexts. A dynamic branch fusion mechanism integrates these branches to optimally aggregate their complementary views. Extensive experiments on five real-world datasets demonstrate that MSSTAN achieves state-of-the-art or highly competitive performance, validating its efficacy for traffic forecasting.

Multi-party urban flow analysis is a crucial task in smart cities. However, existing analysis methods has difficulty in trade-off between data privacy security and spatio-temporal feature capture. The solution to the problem of how to capture the complete spatio-temporal features of multi-party urban flow data while protecting data privacy is of great importance in multi-party urban flow analysis. Therefore, to address data privacy and spatio-temporal feature capture in multi-party urban flow analysis, this paper proposes a spatio-temporal federated analysis model, for multi-party urban flow mining, which is able to effectively protect data privacy and capture spatio-temporal features completely at the same time. Firstly, a multi-party urban flow mining framework based on federated learning is proposed to realize complete capture of spatio-temporal feature information of multi-party urban flow data and mining urban flow pattern knowledge under the premise of protecting data privacy. Secondly, to address the communication cost of the multi-party urban flow analysis, we propose a lazy aggregation method based on similarity clustering, which improves the communication efficiency between clients and the server. Further, we propose a similarity evaluation criteria for urban flow data based on step function, which can effectively calculate the similarity between urban flow data. Finally, we compare the proposed model with some benchmark methods on Chengdu Didi order data and point of interest data to prove the effectiveness of the proposed model and visualize and analyze the spatio-temporal features.

Time series forecasting has become a critical task in data engineering, with the volume of time series data projected to reach 180 ZB by 2025. While traditional forecasting models are typically constrained to single domains, missing opportunities for transferring temporal patterns across different domains. Through analysis, we observe that time series from different domains, despite their distinct statistical characteristics, can be fundamentally understood through temporal dependency patterns, which manifest as either long-term dependencies ( like trends and cycles) or short-term dependencies ( like fluctuations and abrupt changes). This observation motivates us to rethink cross-domain modeling from the dependency preferences perspective. We propose LSTPO, a novel framework that captures cross-domain commonalities through temporal dependency preferences and leverages a meta-learning-based approach to prevent cross-domain training forgetting. LSTPO dynamically models changes in preference over time and swiftly adapts to preference variations across different domains, enabling robust cross-domain forecasting. Through extensive experimental evaluations, we have shown that LSTPO substantially outperforms state-of-the-art forecasting methods while enhancing model transferability under few-shot learning conditions. The source code will be made publicly available upon acceptance.

Unsupervised feature selection aims to enhance the quality of unlabeled data, thereby improving the performance of subsequent unsupervised learning models. However, most of the existing unsupervised feature selection methods rely on single-granularity modeling, which reduces the expressive capability of data to some extent. In addition, the existing studies are generally based on a forward greedy feature selection strategy, which tends to fall into a local optimum. To address these issues, this paper proposes a novel unsupervised feature selection method for handling hybrid data, called unsupervised feature selection method using fuzzy graph momentum random walk in bi-level granular-ball knowledge space. Specifically, a Bi-level Granular-ball Knowledge Space (BGKS) is first constructed by combining fine granularity and coarse granularity representations through a hybrid Gaussian kernel function. Then, a multi-granularity fuzzy graph is built on the BGKS using upper and lower fuzzy approximation operators. Based on this graph, a Momentum Random Walk (MRW) mechanism is introduced to design the Fuzzy Graph Momentum Random Walk (FGMRW) model. Finally, an iterative unsupervised feature selection algorithm is developed. Extensive experiments on 20 public datasets demonstrate that, compared with existing algorithms, the proposed method is able to maintain or even improve clustering performance while selecting fewer features, thus achieving superior overall performance. The source code of this work is publicly available athttps://github.com/HongtaoGao-code/FGMRW-UFS.

Multi-view unsupervised feature selection (MV-UFS) has recently gained significant attention as an effective technique for handling high-dimensional multi-view data. Despite recent progress, three flaws are present in most existing MV-UFS methods. First, they predominantly pay attention to the feature dimensionality while neglecting sample size considerations, typically resulting in cubic time complexity concerning sample size (i.e.$\mathcal {O}(n^{3})$) that hinders them in real-world applications. Second, existing methods primarily focus on the relationships between pairwise views, failing to leverage the complex interactions among multiple views effectively. Third, they generally overlook crucial graph semantic information during graph fusion. An ingenious MV-UFS method is presented in this paper, namedJoint Information Interaction and Semantic Fusion for Multi-View Unsupervised Feature Selection (JIISF), to overcome these limitations. Specifically, JIISF devises a multi-view double reduction learning framework that projects original data into a low-dimensional embedding space and dynamically picks out representative anchors, eliminating redundant information while constructing compact view-specific anchor graphs. Meanwhile, JIISF presents a cross-view information interaction scheme, which enables comprehensive information interaction among multiple views rather than being limited to pairwise views. A semantic alignment fusion strategy is presented to learn a diverse, consistent anchor graph with an explicit semantic alignment module, which can preserve diversity while maintaining consistency. Finally, an alternating optimization algorithm is devised to resolve the proposed JIISF, and experimental results on real-life datasets validate its advantage compared with some state-of-the-art UFS and MV-UFS methods.

Hashing techniques have attracted significant attention in cross-modal retrieval tasks due to their high query efficiency and low storage overhead. However, some methods map heterogeneous features and labels into the same semantic space, ignoring their complementary relationship, which leads to semantic loss. When leveraging pairwise similarity relationships, the affinity matrix incurs high computational costs, and the asymmetric proxy strategy lacks sufficient semantic degrading retrieval performance. Additionally, optimizing binary hash codes remains a challenging task. A novel algorithm, Multi-granularity Semantic Complementarity Fusion Hashing (MSCFH), is proposed to address these issues. First, it learns multi-granularity semantic representations from features and labels, then complements their semantics through a complementarity fusion strategy. Subsequently, an asymmetric semantic learning framework is constructed, guided by label similarities, to learn discriminative hash codes. Furthermore, an efficient discrete optimization algorithm is introduced to update the hash codes. Extensive experiments on three benchmark datasets show that the proposed MSCFH demonstrates the best retrieval accuracy and highly competitive training efficiency compared to state-of-the-art approaches.

Multimodal sentiment analysis aims to comprehensively understand human sentiment by integrating diverse modalities, such as text, audio, and vision. To improve modality complementarity, the recent Multimodal Multi-task Learning (MML) framework employs joint training of unimodal and multimodal sentiment analysis tasks using sub-annotations of modality. In this work, we further draw attention to the observation that integrating unimodal tasks may introduce conflicting task information, negatively affecting the multimodal task performance. Motivated by this issue, we propose the Multimodal Task Correlation-aware Learning (MTCL) framework to leverage beneficial task correlations and suppress harmful ones. Specifically, MTCL introduces a Correlation-Adaptive Training (CAT) strategy to learn a task-relation aware unimodal encoder for each modality. First, in order to distinguish whether a sample contains conflicting information, CAT strategy incorporates a Dual-Branch Contrast (DBC) module which divides the training set into a beneficial subset and a harmful subset. Based on this division, CAT strategy proposes an adaptive training loss to guide the model in understanding nuanced multitask correlations. The adaptive training loss has two components: (1) For the beneficial subset, a contrastive loss is utilized to improve the model’s ability to extract complementary representations. (2) For the harmful subset, we apply a task-correction loss to mitigate the negative interference caused by harmful task associations. With the CAT strategy, our framework can effectively distinguish beneficial and harmful task correlations to extract distinctive and robust unimodal representations. The superiority of MTCL is verified via extensive experiments on several multimodal video sentiment analysis benchmarks. Our work is publicly available at https://github.com/tiggers23/MTCL .

Multi-view unsupervised feature selection (MUFS) has received considerable attention in recent years. Existing MUFS methods for processing unlabeled incomplete multi-view data, where some samples are missing in certain views, first impute the missing values and then perform feature selection on the completed dataset. However, treating imputation and feature selection as two separate processes overlooks their potential interactions. The graph-guided local structure gleaned from feature selection can aid in imputation, which in turn can enhance the feature selection performance. Additionally, most similarity graph-based MUFS methods suffer from high computational costs. To address these problems, we propose a novel MUFS method, termed Tensorial Incomplete Multi-view unsupErvised Feature Selection (TIME-FS). TIME-FS unifies missing value recovery, discriminative feature selection, and low-dimensional representation learning within a joint framework through matrix decomposition. Then, TIME-FS conducts CP decomposition on tensor data formed by the low-dimensional representations of different views to learn a consistent anchor graph across views and a view-preference weight matrix, both of which simultaneously guide missing view imputation and feature selection. Furthermore, an efficient algorithm with low time complexity and rapid convergence is proposed to solve TIME-FS. Extensive experimental results demonstrate the effectiveness and efficiency of TIME-FS over state-of-the-art methods.

Open intent classification is critical for the development of dialogue systems, aiming to accurately classify known intents into their corresponding classes while identifying unknown intents. Prior boundary-based methods assumed known intents fit within compact spherical regions, focusing on coarse-grained representation and precise spherical decision boundaries. However, these assumptions are often violated in practical scenarios, making it difficult to distinguish known intent classes from unknowns using a single spherical boundary. To tackle these issues, we propose a Multi-granularity Open intent classification method via adaptive Granular-Ball decision boundary (MOGB). Our MOGB method consists of two modules: representation learning and decision boundary acquiring. To effectively represent the intent distribution, we design a hierarchical representation learning method. This involves iteratively alternating between adaptive granular-ball clustering and nearest sub-centroid classification to capture fine-grained semantic structures within known intent classes. Furthermore, multi-granularity decision boundaries are constructed for open intent classification by employing granular-balls with varying centroids and radii. Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method.

Understanding how humans cooperatively utilize semantic knowledge to explore unfamiliar environments and decide on navigation directions is critical for house service multi-robot systems. Previous methods primarily focused on single-robot centralized planning strategies, which severely limited exploration efficiency. Recent research has considered decentralized planning strategies for multiple robots, assigning separate planning models to each robot, but these approaches often overlook communication costs. In this work, we propose Multimodal Chain-of-Thought Co-Navigation (MCoCoNav), a modular approach that utilizes multimodal Chain-of-Thought to plan collaborative semantic navigation for multiple robots. MCoCoNav combines visual perception with Vision Language Models (VLMs) to evaluate exploration value through probabilistic scoring, thus reducing time costs and achieving stable outputs. Additionally, a global semantic map is used as a communication bridge, minimizing communication overhead while integrating observational results. Guided by scores that reflect exploration trends, robots utilize this map to assess whether to explore new frontier points or revisit history nodes. Experiments on HM3D_v0.2 and MP3D demonstrate the effectiveness of our approach.

Magnetic resonance imaging (MRI) is essential for studying neurological and psychiatric disorders, but its analysis is limited by sparse connectivity and individual variability. Most existing methods rely on pairwise Pearson correlation, which captures only direct interactions and overlooks indirect, mediated connections, leading to oversimplified brain networks. To address this, we propose Dynamic Graph Enhancement -Multi-level Attention Weighting (DGE-MAW), a data-centered approach that improves structural representation and diagnostic interpretability. It comprises two main modules: Dynamic Graph Enhancement dynamically updates the graph structure to identify indirect connections. By integrating intermediate nodes, nonlinear activations, and topological indicators, it mimics neural transmission pathways and reconstructs more informative brain networks, reducing information loss. Multi-level Attention Weighting introduces a hierarchical attention mechanism inspired by biological signal transmission, capturing individual, regional, and global semantics. This enables interpretable and adaptive integration of multimodal features. Experiments show that DGE-MAW significantly enhances data quality, interpretability, multimodal fusion, and diagnostic accuracy. On the ABIDE and ADHD-200 datasets, it outperforms state-of-the-art model-centric methods across various neuroimaging tasks. By modeling indirect connections and incorporating prior brain knowledge, it demonstrates the potential of MRI augmentation. Code and resources are available at: https://github.com/A3-247/DGE-MAW.

Class Incremental Learning (CIL) aims to enable models to learn new classes sequentially while retaining knowledge of previous ones. Although current methods have alleviated catastrophic forgetting (CF), recent studies highlight that the performance of CIL models is highly sensitive to the order of class arrival, particularly when sequentially introduced classes exhibit high inter-class similarity. To address this critical yet understudied challenge of class order sensitivity, we first extend existing CIL frameworks through theoretical analysis, proving that grouping classes with lower pairwise similarity during incremental phases significantly improves model robustness to order variations. Building on this insight, we propose Graph-Driven Dynamic Similarity Grouping (GDDSG), a novel method that employs graph coloring algorithms to dynamically partition classes into similarity-constrained groups. Each group trains an isolated CIL sub-model and constructs meta-features for class group identification. Experimental results demonstrate that our method effectively addresses the issue of class order sensitivity while achieving optimal performance in both model accuracy and anti-forgetting capability. Our code is available at https://github.com/AIGNLAI/GDDSG.

Federated continual learning (FCL) allows each client to continually update its knowledge from task streams, enhancing the applicability of federated learning in real-world scenarios. However, FCL needs to address not only spatial data heterogeneity between clients but also temporal data heterogeneity between tasks. In this paper, empirical experiments demonstrate that such input-level heterogeneity significantly affects the model’s internal parameters and outputs, leading to severe spatial-temporal catastrophic forgetting of local and previous knowledge. To this end, we propose Federated Tail Anchor (FedTA) to mix trainable Tail Anchor with the frozen output features to adjust their position in the feature space, thereby overcoming parameter-forgetting and output-forgetting. Three novel components are also included: Input Enhancement for improving the performance of pre-trained models on downstream tasks; Selective Input Knowledge Fusion for fusion of heterogeneous local knowledge on the server; and Best Global Prototype Selection for finding the best anchor point for each class in the feature space. Extensive experiments demonstrate that FedTA not only outperforms existing FCL methods but also effectively preserves the relative positions of features. Code is available at: https://github.com/SkyOfBeginning/FedTA_CVPR2025.

Reward acts as a signal to guide the agent’s learning process in Reinforcement Learning (RL), evaluating and assigning rewards to the agent’s actions based on theiralignment with goals. Designing reward is challenging in multiagent environment such as StarCraft II benchmark since agents face credit allocation and role adaptation problems. Recent studies have successfully exploited the language understanding and reasoning capabilities of large language models (LLMs) to learn manipulation tasks. Impressed by the remarkable power of LLMs, this paper employs LLMs as role-specific reward designer for playing StarCraft II, making rewards more flexible and task-oriented. Firstly, we develop an interactive text and multiagent RL environment to study real-time strategy generation in StarCraft II. Secondly, we use LLMs to interpret the game situation and understand agent roles from user instructions. Then, by assigning appropriate subtasks, LLMs quantify the completion of these subtasks to generate role-specific rewards. Further, credit assignment problem is addressed by introducing dynamic reward weights in value decomposition method. In StarCraft II maps, experiments show that role-aligned RL agents trained with our framework achieve superior policy performance, and win rate results demonstrates the effectiveness of our approach in decision-making for micromanagement and long-term planning.

Improving the performance of single image super-resolution (SISR) via extending the effective receptive field (ERF) of the model has become an admired paradigm in the field due to the universal self-similarity prior of natural images. However, it cannot fully explore model capability by solely increasing the ERF to capture long-range dependencies as the non-local self-similarity is typically multi-scale and cross-scale. To this end, a Long-range Multi-scale Fusion Network (LMFN) is devised in this work to simultaneously excavate both long-range and multi-scale priors in images, and the interaction between the both. Within the same scale, our model employs large kernel attention (LKA) and multi-scale modulation (MSM) to learn long-range and multi-scale features. To exploit the interaction between long-range and multi-scale dependencies within one single scale and across scales, we design an Interactive Fusion Modulation (IFM) module for the effective fusion of the non-local and multi-scale features. Extensive experiments on the benchmark datasets illustrate the significant superiority of the proposed LMFN over the advanced SISR models.

Embodied Question Answering (EQA) is a task in artificial intelligence where an intelligent agent is required to answer questions about its environment. For example, to answer a question such as "Is the TV on or off?", the agent must navigate to the room with the TV and answer with either "On." or "Off." after recognizing the status. Unlike traditional question-answering systems that rely solely on text or static images, EQA involves agents that can move through a physical or simulated space, interact with the environment, and gather information to respond accurately. The agent must interpret both visual and linguistic inputs, navigate the environment, and complete tasks or locate objects based on the user’s questions. However, in the real world, the agent always faces unseen environments (i.e. different people’s houses), which makes the pre-trained model fail. Meanwhile, re-training in an unseen environment can cause high costs. Therefore, it is significant for the agent to learn continually by itself to cope with the challenges of unseen environments. In this work, we proposed a continual learning method based on generative adversarial imitation learning and self-supervision to support the agent when facing unseen environments. Besides, we designed a policy generator and policy quality discriminator to generate action policy sequences and evaluate the quality of the policy, respectively. Extensive experiments on the MP3D-EQA dataset demonstrate that our method reaches state-of-the-art performance.

In the scope of 3D human pose estimation, the task encompasses estimating the 3D positions of key skeletal points (i.e., wrists, elbows, and knees) from a 2D image or video sequence. This technology demonstrates widespread applicability across diverse domains, encompassing domains such as kinematic analysis, virtual reality, augmented reality, and medical imaging analysis. The common approach is divided into two stages: i) 2D Keypoint Detection: Detecting 2D keypoints from images. ii) 2D-to-3D Lifting: Converting 2D keypoints into 3D coordinates. Present research predominantly concentrates on Stage 2 and leverages sophisticated deep learning architectures, notably Transformers, yet two key challenges exist: i) suboptimal performance in predicting local details; ii) susceptibility to noise interference. In this work, we propose a time-principal component fusion model with limb segment property tracking is proposed to address the challenges above. By integrating timedomain features with principal components extracted through the Karhunen-Loève Transform (KLT), the model aims to address challenges related to feature extraction and noise reduction. Furthermore, to address the issue of suboptimal performance in predicting local details, we devise a property transformer to track the lengths of limb segments and predict the fixed property. Extensive experiments demonstrate that KLFormer showcases state-of-the-art performance on the standard benchmark dataset, Human3.6M.

Out-of-distribution (OOD) detection in 3D point cloud data remains a challenge, particularly in applications where safe and robust perception is critical. While existing OOD detection methods have shown progress for 2D image data, extending these to 3D environments involves unique obstacles. This paper introduces a training-free framework that leverages Vision-Language Models (VLMs) for effective OOD detection in 3D point clouds. By constructing a graph based on class prototypes and testing data, we exploit the data manifold structure to enhancing the effectiveness of VLMs for 3D OOD detection. We propose a novel Graph Score Propagation (GSP) method that incorporates prompt clustering and self-training negative prompting to improve OOD scoring with VLM. Our method is also adaptable to few-shot scenarios, providing options for practical applications. We demonstrate that GSP consistently outperforms state-of-the-art methods across synthetic and real-world datasets 3D point cloud OOD detection.

Instance ImageGoal Navigation (IIN) entails an agent autonomously seeking out a specific object instance depicted by a goal image in an unknown environment. While Large Language Models (LLMs) have shown promise in navigation tasks similar to IIN, their application to IIN remains unexplored. Furthermore, existing LLM-based exploration faces challenges such as inaccurate reasoning due to informative environmental information available to the agent, especially in the early episode stages, and the inability of reinforcement learning(RL) exploration to fully leverage gathered information. Moreover, previous IIN methods did not productively verify potentially distant goal objects discovered during exploration. This work proposes MoPE–Mixture of Policy Experts for exploration and potential goal verification with multimodal information when exploring. Specifically, the hybrid exploration policy comprises an LLM and an RL-based Policy Network (RLPN) to generate an exploration goal to explore efficiently. Our MoPE model surpasses prior approaches on the HM3D datasets significantly.

Deep learning-based approaches have demonstrated significant advancements in time series forecasting. Despite these ongoing developments, the complex dynamics of time series make it challenging to establish the rule of thumb for designing the golden model architecture. In this study, we argue that refining existing advanced models through a universal calibrating strategy can deliver substantial benefits with minimal resource costs, as opposed to elaborating and training a new model from scratch. We first identify a multi-target learning conflict in the calibrating process, which arises when optimizing variables across time steps, leading to the underutilization of the model's learning capabilities. To address this issue, we propose an innovative calibrating strategy called Socket+Plug (SoP). This approach retains an exclusive optimizer and early-stopping monitor for each predicted target within each Plug while keeping the fully trained Socket backbone frozen. The model-agnostic nature of SoP allows it to directly calibrate the performance of any trained deep forecasting models, regardless of their specific architectures. Extensive experiments on various time series benchmarks and a spatio-temporal meteorological ERA5 dataset demonstrate the effectiveness of SoP, achieving up to a 22% improvement even when employing a simple MLP as the Plug (highlighted in Figure 1).

Established sampling protocols for 3D point cloud learning, such as Farthest Point Sampling (FPS) and Fixed Sample Size (FSS), have long been relied upon. However, real-world data often suffer from corruptions, such as sensor noise, which violates the benign data assumption in current protocols. As a result, these protocols are highly vulnerable to noise, posing significant safety risks in critical applications like autonomous driving. To address these issues, we propose an enhanced point cloud sampling protocol, PointSP, designed to improve robustness against point cloud corruptions. PointSP incorporates key point reweighting to mitigate outlier sensitivity and ensure the selection of representative points. It also introduces a local-global balanced downsampling strategy, which allows for scalable and adaptive sampling while maintaining geometric consistency. Additionally, a lightweight tangent plane interpolation method is used to preserve local geometry while enhancing the density of the point cloud. Unlike learning-based approaches that require additional model training, PointSP is architecture-agnostic, requiring no extra learning or modification to the network. This enables seamless integration into existing pipelines. Extensive experiments on synthetic and real-world corrupted datasets show that PointSP significantly improves the robustness and accuracy of point cloud classification, outperforming state-of-the-art methods across multiple benchmarks.

In the realm of dermatology, Convolutional Neural Networks (CNNs) have exhibited immense potential, emerging as vital tools to aid physicians for precise diagnosis. However, current work generally faces two main challenges: (1) the inadequate samples in dermatological image datasets hinder sufficiently training high-performance deep models; (2) the pronounced issue of class imbalance significantly undermines the models’ ability to recognize minority class conditions. In response to these problems, this work introduces a novel network architecture and loss called Hierarchical Feature Learning (HFL), aimed at enhancing the model’s feature representation capabilities and adaptability to imbalanced data. Specifically, we devise a CNN architecture based on hierarchical feature storage, which effectively augments the network’s proficiency in extracting and preserving rich, hierarchical features through the incorporation of a multi-level feature memory mechanism. To further mitigate the impact of data imbalance on model training, we propose an Adaptive Re-weighted Loss (ARL). This loss dynamically adjusts the loss weights according to the real-time distribution of samples across classes, ensuring that minority class samples receive adequate attention during training. Experimental results on ISIC-2017, ISIC-2018 and 7-PT datasets demonstrate the effectiveness of our proposed methodology.

Continual Learning (CL) primarily aims to retain knowledge to prevent catastrophic forgetting and transfer knowledge to facilitate learning new tasks. Unlike traditional methods, we propose a novel perspective: CL not only needs to prevent forgetting, but also requires intentional forgetting.This arises from existing CL methods ignoring biases in real-world data, leading the model to learn spurious correlations that transfer and amplify across tasks. From feature extraction and prediction results, we find that data biases simultaneously reduce CL's ability to retain and transfer knowledge. To address this, we propose ErrorEraser, a universal plugin that removes erroneous memories caused by biases in CL, enhancing performance in both new and old tasks. ErrorEraser consists of two modules: Error Identification and Error Erasure. The former learns the probability density distribution of task data in the feature space without prior knowledge, enabling accurate identification of potentially biased samples. The latter ensures only erroneous knowledge is erased by shifting the decision space of representative outlier samples. Additionally, an incremental feature distribution learning strategy is designed to reduce the resource overhead during error identification in downstream tasks. Extensive experimental results show that ErrorEraser significantly mitigates the negative impact of data biases, achieving higher accuracy and lower forgetting rates across three types of CL methods. The code is available at https://github.com/diadai/ErrorEraser.

Open-world continual learning (OWCL) adapts to sequential tasks with open samples, learning knowledge incrementally while preventing forgetting. However, existing OWCL still requires a large amount of labeled data for training, which is often impractical in real-world applications. Given that new categories/entities typically come with limited annotations and are in small quantities, a more realistic situation is OWCL with scarce labeled data, i.e., few-shot training samples. Hence, this paper investigates the problem of open-world few-shot continual learning (OFCL), challenging in (i) learning unbounded tasks without forgetting previous knowledge and avoiding overfitting(ii) constructing compact decision boundaries for open detection with limited labeled data, and (iii) transferring knowledge about knowns and unknowns and even update the unknowns to knowns once the labels of open samples are learned. In response, we propose a novel OFCL framework that integrates three key components: (1) an instance-wise token augmentation (ITA) that represents and enriches sample representations with additional knowledge(2) a margin-based open boundary (MOB) that supports open detection with new tasks emerge over time, and (3) an adaptive knowledge space (AKS) that endows unknowns with knowledge for the updating from unknowns to knowns. Finally, extensive experiments show that the proposed OFCL framework outperforms all baselines remarkably with practical importance and reproducibility. The source code is released at https://github.com/liyj1201/OFCL.

Multivariate Time Series Forecasting (MTSF) has long been a key research focus. Traditionally, these studies assume a fixed number of variables, but in real-world applications, Cyber-Physical Systems often expand as new sensors are deployed, increasing variables in MTSF. In light of this, we introduce a novel task, Expanding-variate Time Series Forecasting (EVTSF). This task presents unique challenges, specifically (1) handling inconsistent data shapes caused by adding new variables, and (2) addressing imbalanced spatio-temporal learning, where expanding variables have limited observed data due to the necessity for timely operation. To address these challenges, we propose STEV, a flexible spatio-temporal forecasting framework. STEV includes a new Flat Scheme to tackle the inconsistent data shape issue, which extends the graph-based spatio-temporal modeling architecture into 1D space by flattening the 2D samples along the variable dimension, making the model variable-scale-agnostic while still preserving dynamic spatial correlations through a holistic graph. Additionally, we introduce a novel Spatio-temporal Focal Learning strategy that incorporates a negative filter to resolve potential conflicts between contrastive learning and graph representation, and a focal contrastive loss as its core to guide the framework to focus on optimizing the expanding variables. To evaluate the effectiveness of STEV, we benchmark EVTSF performance on three real-world datasets from various domains and compare it against three potential solutions employing state-of-the-art (SOTA) MTSF models tailored for EVSTF. Experimental results show that STEV significantly outperforms its competitors, especially in handling expanding variables. Notably, STEV, with only 5% of observations during the expanding period, is on par with SOTA MTSF models trained with complete data. Further exploration of various expanding scenarios underscores the generalizability of STEV in real-world applications.

While existing research on Multilingual CLIP (MCLIP) has prioritized model architecture design, our work uncovers a critical challenge in practical adaptation: fine-tuning MCLIP through a single source language risks diminishing its multilingual capabilities in downstream tasks due to cross-linguistic disparities. To bridge this gap, we systematically investigate the role of ‌token similarity‌ in cross-lingual transferability for image-text retrieval, establishing it as a key factor governing fine-tuning efficacy. Building on this insight, we propose two novel strategies to enhance efficiency while preserving multilinguality: 1) ‌TaPCL dynamically optimizes training by prioritizing linguistically distant language pairs during corpus sampling, reducing redundant computation, and 2) CiPCL enriches the source corpus with multilingual key terms, enabling targeted knowledge transfer without reliance on exhaustive parallel data. By strategically balancing token similarity and domain-critical information, our methods significantly lower computational costs and mitigate over-dependence on parallel corpora. Experimental evaluations across diverse datasets validate the effectiveness and scalability of our framework, demonstrating robust multilingual retention across languages. This work provides a principled pathway for adapting MCLIP to real-world scenarios, where computational efficiency and cross-lingual robustness are paramount. Our codes are available at https://github.com/tiggers23/TaPCL-CiPCL.

Sarcasm is a subtle expression that indicates the incongruity between literal meanings and factual opinions. For multimodal posts in social medias which consist of both images and texts, sarcasm expressions are even more widespread. Recent works have paid attentions to Multimodal Sarcasm Target Identification (MSTI), which focuses on detecting aspect terms of mockery or ridicule as sarcasm targets. However, the current MSTI benchmark only contains annotations on fine-grained sarcasm targets within sarcastic samples. In practice, it will be featured by two major limitations. First, there lack annotations on non-sarcasm aspects to inform deep models to perceive the semantic difference between sarcasm targets and non-sarcasm aspects. As a result, deep models will tend to incorrectly recognize non-sarcasm aspects as sarcasm targets. Second, there lack non-sarcasm samples to inform deep models to perceive the inherent semantics of sarcasm intentions. Due to the subtle characteristic of sarcasm expressions, models trained with only fine-grained supervision signals cannot thoroughly understand the sarcasm semantics, making the fine-grained task of sarcasm target identification restricted. Motivated by these limitations, this work reconstructs a more comprehensive MSTI benchmark by introducing both fine-grained non-sarcasm aspect annotations for existing sarcasm samples and non-sarcastic samples as non-sarcasm references to enable deep models to clearly perceive the mentioned information during training. Based on the multi-granularity (i.e., both aspect-level and sample-level) non-sarcasm information introduced into this new benchmark, this work further proposes a pluggable Semantics-aware Sarcasm Target Identification mechanism to enhance sarcasm target identification by modeling the overall semantics of sarcasm intentions via an auxiliary sample-level sarcasm recognition task. By modeling the overall semantics of sarcasm intention, deep models can obtain a more comprehensive understanding on sarcasm semantics, leading to improved performance on fine-grained sarcasm target identification. Extensive experiments are conducted to validate our contribution. Both the dataset and code are available at https://github.com/tiggers23/MSTI-Plus.

Graph-based feature selection methods can capture the most discriminative subset of features in multi-label data and have made great strides in recent years. However, most existing methods have three drawbacks: (1) Only binary pairwise relationship is difficult to capture latent structural information in high-dimensional data fully. (2) Using only one distance metric cannot accurately mine the similarity relationship between instances. (3) Ignore the problem that some of the non-obvious labels are missing when labels are abundant. Therefore, this paper proposes a graph diffusion with dual-distance metrics for missing multi-label feature selection named GDMMFS. GDMMFS uses both Euclidean and cosine distances to explore instance correlations and then uses diffusion techniques to transform binary relationship into quaternary relationship, thus obtaining a stable similarity graph. Moreover, GDMMFS utilizes adaptive graphs to learn latent structure information to improve the model’s immunity to interference. Finally, the ℓ 2 , 1 -norm is used as a sparse constraint to guide the weight matrix to capture the most discriminative subset of features. To further consider the possibility of missing label data, GDMMFS constructed a logic matrix to guide the numerical labels in recovering the missing information. Comparison experiments with state-of-the-art related algorithms demonstrate the superiority of the proposed algorithm, and ablation experiments demonstrate the effectiveness of graph diffusion with two-distance metrics.

Air quality modeling and forecasting has become a key problem in environmental protection. The existing prediction models typically require large‐scale and high‐quality historical data to achieve better performance. However, insufficient data volume and significant differences between data distribution across different regions will definitely reduce the effectiveness of the model reuse. To address the above issues, we propose a novel hybrid recurrent network based on domain adversarial transfer to achieve a stronger generalization ability when training air quality data from multisource domains. The proposed model mainly consists of three fundamental modules, i.e., feature extractor, regression predictor, and domain classifier. One‐dimensional convolutional neural networks (1D‐CNNs) are used to extract temporal feature of data from source and target stations. Bi‐directional gated recurrent unit (bi‐GRU) and bi‐directional long short‐term memory (bi‐LSTM) are utilized to learn temporal dependencies pattern of multivariate time series data. Two adversarial transfer strategies are employed to ensure that our model is capable of finding domain invariant representations automatically. Experiments with different number of source domains are conducted to demonstrate the effectiveness of the proposed domain transfer strategies. The experimental results also show that our composite model has superior performance for forecasting air quality in various regions. As further evidence, the adversarial training method could promote the positive transfer and alleviate the negative effect of irrelevant source data. Besides, our model exhibits preferable generalization capability as more robust prediction results are achieved on both unseen target domains and original source domains.

Air quality prediction utilizes IoT technologies to collect data centrally for model training, which may cause regulatory risks, privacy concerns, and high costs of integrating data. Meanwhile, distributed training through federated learning relies on the pre-defined graph structure generated by the geographic locations of air monitoring stations, which fails to capture the potential spatial relationships between air monitoring stations. In addition, the inherent multi-period attribute of air quality data makes time series changes extremely complex. To this end, this paper proposes a Federated framework for air quality prediction with Pre-defined graph and Adaptive Graph (FedPAG). Specifically, the client encodes the air quality data and meteorological data provided by the Internet of Things (IoT) system using the multi-period interaction and encoder modules to capture the proximity and periodicity features of the time series data. Next, the server combines the hidden states uploaded by the clients with pre-defined graph and adaptive graph respectively and forms node embeddings to capture the spatial features among the clients. Then, the client concatenates the hidden states with node embeddings to fuse the spatial information and feeds them into the decoder to obtain the final predicted values. Finally, we conduct experiments on the Beijing and Shijiazhuang datasets to demonstrate the effectiveness of the proposed method.

Feature selection (FS) is essential in machine learning and data mining as it makes handling high-dimensional data more efficient and reliable. More attention has been paid to unsupervised feature selection (UFS) due to the extra resources required to obtain labels for data in the real world. Most of the existing embedded UFS utilize a sparse projection matrix for FS. However, this may introduce additional regularization terms, and it is difficult to control the sparsity of the projection matrix well. Moreover, such methods may seriously destroy the original feature structure in the embedding space. Instead, avoiding projecting the original data into the low-dimensional embedding space and identifying features directly from the raw features that perform well in the process of making the data show a distinct cluster structure is a feasible solution. Inspired by this, this paper proposes a model called A General Adaptive Unsupervised Feature Selection with Auto-weighting (GAWFS), which utilizes two techniques, non-negative matrix factorization, and adaptive graph learning, to simulate the process of dividing data into clusters, and identifies the features that are most discriminative in the clustering process by a feature weighting matrix Θ. Since the weighting matrix is sparse, it also plays the role of FS or a filter. Finally, experiments comparing GAWFS with several state-of-the-art UFS methods on synthetic datasets and real-world datasets are conducted, and the results demonstrate the superiority of the GAWFS.

Traffic flow forecasting task plays an essential role in intelligent transportation systems. Accurately capturing the intricate spatio-temporal dependencies in traffic network signals is the core of precise prediction. Recently, a paradigm that models spatio-temporal dependencies through graph neural networks and time series models has become one of the most promising methods to solve this problem. However, existing methods still have limitations due to ineffectively modeling dynamic spatial dependencies and high time and space complexity. To address these issues, we propose a simplifying and powerful general spatio-temporal traffic flow forecasting model called LightST. Specifically, LightST first embeds temporal covariates and spatial position information to enhance the spatio-temporal modeling capabilities. Then, stacked temporal linear layers are introduced to capture temporal dependencies efficiently. Finally,we propose a concise adaptive spatio-temporal embedding graph convolution method to extract implicit spatial dependencies over time via dynamic graph convolution with adaptive spatio-temporal embedding graph generation. Extensive experiment results on four public traffic flow datasets demonstrate the superiority of our LightST concerning computational efficiency and prediction performance.

Multiview data possess different discriminability in different views, which is challenging to catch but crucial for a feature selection model. Multiscale information, which represents vertical exploration in each view, is vital for further mining traits implied in multiview data. However, most existing studies neglect these beneficial multi-granulation characteristics. This study first embeds the multiscale information into the sparse learning framework for multiview feature selection. A class-specific discriminability and multiscale information-based multiview feature selection (CDMIMFS) method is proposed. It explores the fuzzy and uncertain class-specific discriminability which is inherently discrepant in different views by the fuzzy rough set theory. It relaxes the over strict requirement for complete consistency in multiscale information systems to make a trade-off, which further enhances discriminative feature selection. An effective iteration algorithm is proposed to solve the optimization. Both the theoretical proof and experimental demonstration of convergence are provided. Comprehensive experiments are conducted on the CDMIMFS compared with state-of-the-art algorithms. Results on different evaluation metrics exhibit the advantages of the proposed method.

Image super-resolution (SR) has significantly advanced through the adoption of Transformer architectures. However, conventional techniques aimed at enlarging the self-attention window to capture broader contexts come with inherent drawbacks, especially the significantly increased computational demands. Moreover, the feature perception within a fixed-size window of existing models restricts the effective receptive field (ERF) and the intermediate feature diversity. We demonstrate that a flexible integration of attention across diverse spatial extents can yield significant performance enhancements. In line with this insight, we introduce Multi-Range Attention Transformer (MAT) for SR tasks. MAT leverages the computational advantages inherent in dilation operation, in conjunction with self-attention mechanism, to facilitate both multi-range attention (MA) and sparse multi-range attention (SMA), enabling efficient capture of both regional and sparse global features. Combined with local feature extraction, MAT adeptly capture dependencies across various spatial ranges, improving the diversity and efficacy of its feature representations. We also introduce the MSConvStar module, which augments the model’s ability for multi-range representation learning. Comprehensive experiments show that our MAT exhibits superior performance to existing state-of-the-art SR models with remarkable efficiency ($\sim 3.3\times $faster than SRFormer-light). The codes are available athttps://github.com/stella-von/MAT.

The inherent characteristics involved in data can be mined from multi-scale information systems by extracting information from different value levels of features. In real applications, noise data and irrelevant or redundant features affect the generality of learning models. Therefore, keeping meaningful features and avoiding the effect of noise is essential for feature selection in a multi-scale information system. In bireduct, multi-scale granulation can be used to characterize the importance and correlation of features at different scales. However, little work has taken the distribution of multi-scale data into account when granulating it. In addition, these approaches focus on solving the task of multi-scale data reduction only from the dimension perspective. To this end, a fuzzy-rough bireduct with supervised multi-scale granulation (FrBSmg) is proposed. First, the supervised multi-scale fuzzy granulation based on data distribution is constructed. Then, scaled uncertainty measures are defined to describe the fuzzy relevance of each feature. Furthermore, the global and local distributions of a sample are characterized simultaneously based on the positive region, which can reflect the degree of a sample belonging to some class, and the supervised fuzzy similarity relation can describe the degree of a sample belonging to its class. A strategy of Feature-Correlated Selection and Sample-Noisy Removal is devised for bireduct. Finally, the experimental results on twenty-one public datasets show the effectiveness of FrBSmg.

In the realm of data mining, outlier detection has emerged as a pivotal research focus, aimed at uncovering anomalies within datasets to extract meaningful and valuable insights. The objective is to leverage data mining methodologies to pinpoint anomalies within datasets, thereby revealing crucial and enlightening information. Herein, we introduce a groundbreaking outlier detection methodology, ODMGIS, that seamlessly integrates multigranularity representation and information set concepts to devise the multigranularity information set (MGIS) model. This model adeptly characterizes the distribution patterns of data points. First, we employ entropy function and their complementary function as measurement tools to accurately quantify the inherent uncertainty in data with different distributions, and considers the sum of the two as a comprehensive representation of the overall uncertainty of the information source. Subsequently, an outlier score model is constructed based on MGIS, which can deeply characterize the degree of outlierness of samples, thereby effectively identifying abnormal points in the dataset. During validation, ODMGIS was rigorously tested on practical datasets from medicine and bioinformatics, and its performance was benchmarked against both traditional and the state-of-the-art algorithms, showcasing substantial benefits. This research not only contributes a fresh perspective to outlier detection, but also sparks innovative avenues for exploring and advancing the granular computing theory.

Co-representation, which co-represents samples and features, has been widely used in various machine learning tasks, such as document clustering, gene expression analysis, and recommendation systems. It not only reveals the cluster structure of both samples and features, but also reveals the sample–feature correlation. Given a tabular data matrix, co-representation usually exhibits as the co-occurrence structures of rows and columns. However, identifying such structured patterns in complex real-world data can be very challenging. To address this problem, we propose an unsupervised discriminative co-representation learning model based on multidimensional scaling (DCLMDS). The main novelty is that DCLMDS introduces a co-representation learning term to ensure the discriminability between co-occurrence structures. As a result, the co-representation learned by DCLMDS contains richer information of the underlying correlation between samples and features within data. This could subsequently enhance the capacity of machines and systems for processing complex real-world information more proficiently. Furthermore, inspired by the fuzzy set theory, we integrate fuzzy membership degree that can accurately capture the uncertainty within data, thus enabling DCLMDS to learn a more effective co-representation in a soft manner. To evaluate the performance of DCLMDS, we conduct extensive experiments on 18 datasets, and the results demonstrate that DCLMDS can generate both accurate and discriminative co-representation, which well meets our desired outcomes.

Temporal data analysis plays a pivotal role in applications such as weather forecasting, traffic flow management, energy consumption monitoring, and other areas of urban computing. In recent years, temporal data modeling has transitioned from traditional deep learning methods to pre-trained models. However, existing approaches often exhibit significant task-specific limitations, requiring bespoke model designs and extensive domain data for training. To address these challenges, this study introduces KPT, a novel foundation model for temporal data analysis in urban computing. By leveraging temporal competitive attention and feature interaction attention mechanisms, KPT can effectively capture global context, integrate cross-variable features precisely, and achieve universal feature learning across diverse time series tasks. Additionally, the knowledge prompt network facilitates the deep fusion of cross-layer features via an intricate interaction mechanism, enabling the model to identify and align shared temporal patterns across different time series data. These patterns then transformed into knowledge prompts, thereby enhancing the universal feature learning capabilities of the pre-trained model. Experimental results demonstrate that KPT excels in four core temporal analysis tasks within urban computing, outperforming task-specific models. This highlights KPT’s ability to generalize across tasks and underscores its potential as a foundation model for multi-task scenarios in urban computing.

Long-term time series forecasting (LTSF) is a critical task across diverse domains. Despite significant advancements in LTSF research, we identify a performance bottleneck in existing LTSF methods caused by the inadequate modeling of Temporal Dependencies within the Target (TDT). To address this issue, we propose a novel and generic temporal modeling framework, Temporal Dependency Alignment (TDAlign), that equips existing LTSF methods with TDT learning capabilities. TDAlign introduces two key innovations: 1) a loss function that aligns the change values between adjacent time steps in the predictions with those in the target, ensuring consistency with variation patterns, and 2) an adaptive loss balancing strategy that seamlessly integrates the new loss function with existing LTSF methods without introducing additional learnable parameters. As a plug-and-play framework, TDAlign enhances existing methods with minimal computational overhead, featuring only linear time complexity and constant space complexity relative to the prediction length. Extensive experiments on six strong LTSF baselines across seven real-world datasets demonstrate the effectiveness and flexibility of TDAlign. On average, TDAlign reduces baseline prediction errors by1.47%to9.19%and change value errors by4.57%to15.78%, highlighting its substantial performance improvements.

Over the recent years, multimodal named entity recognition has gained increasing attentions due to its wide applications in social media. The key factor of multimodal named entity recognition is to effectively fuse information of different modalities. Existing works mainly focus on reinforcing textual representations by fusing image features via the cross-modal attention mechanism. However, these works are limited in reinforcing the text modality at the token level. As a named entity usually contains several tokens, modeling token-level inter-modal interactions is suboptimal for the multimodal named entity recognition problem. In this work, we propose a multimodal named entity recognition approach dubbed Adaptive Multi-scale Language Reinforcement (AMLR) to implement entity-level language reinforcement. To this end, our model first expands token-level textual representations into multi-scale textual representations which are composed of language units of different lengths. After that, the visual information reinforces the language modality by modeling the cross-modal attention between images and expanded multi-scale textual representations. Unlike existing token-level language reinforcement methods, the word sequences of named entities can be directly interacted with the visual features as a whole, making the modeled cross-modal correlations more reasonable. Although the underlying entity is not given, the training procedure can encourage the relevant image contents to adaptively attend to the appropriate language units, making our approach not rely on the pipeline design. Comprehensive evaluation results on two public Twitter datasets clearly demonstrate the superiority of our proposed model.

Traditional multilabel feature selection (MFS) typically relies on pre-computing global information within the feature space. However, in real-world applications, features are dynamically generated and continuously arrive over time, known as streaming features, rendering many existing approaches ineffective. Some MFS methods for streaming features have been developed, several challenges persist: (1) Previous research often uses certain strategies to model streaming feature evaluation, failing to process fuzzy information effectively; (2) The maximum correlation between features and class is emphasized, while inter-class separability is ignored, leading to inaccurate feature evaluation; (3) The continuous influx of streaming features brings the dynamics and unknowns to data distribution, has been largely overlooked in previous work; (4) Streaming feature selection requires immediate feedback on newly arriving features, posing challenges to the algorithm's real-time responsiveness. Motivated by these observations, this paper introduces a novel online MFS strategy for streaming features. First, the weighted manifold distance is designed, and the fuzzy manifold similarity learning strategy is formalized to analyze the instance relationships of unknown distribution. Second, the fuzzy manifold intra-class correlation and inter-class separability are devised to quantify feature discriminability. Finally, a novel multilabel streaming feature analysis framework is established, with feature discriminability as the guiding factor. Incoming features are categorized as weakly relevant, strongly relevant, or redundant, culminating in generating a reliable feature selection subset. Extensive experiments on fifteen public datasets demonstrate that our algorithm achieves competitive performance compared to nine state-of-the-art offline and online algorithms.

In our daily lives, people frequently consider daily schedule to meet their needs, such as going to a barbershop for a haircut, then eating in a restaurant, and finally shopping in a supermarket. Reasonable activity location [or point-of-interest (POI)] and activity sequencing will help people save a lot of time and get better services. In this article, we propose a reinforcement learning-based deep activity factor balancing model to recommend a reasonable daily schedule according to user's current location and needs. The proposed model consists of a deep activity factor balancing network (DAFB) and a reinforcement learning framework. First, the DAFB is proposed to fuse multiple factors that affect daily schedule recommendation (DSR). Then, a reinforcement learning framework based on policy gradient is used to learn the parameters of the DAFB. Further, on the feature storage based on the matrix method, we compress the feature storage space of the candidate POIs. Finally, the proposed method is compared with seven benchmark methods using two real-world datasets. Experimental results show that the proposed method is adaptive and effective.

Robust knowledge acquisition approaches are one of the research hotspots in data mining. The fuzzy dominance rough sets (FDRS) model is an important knowledge acquisition tool for ordinal classification tasks. However, it has been proved in practice that this model usually performs poorly fault tolerance, and only one noisy sample can cause huge interference in acquiring knowledge. Attribute reduction is one of the nontrivial applications of the FDRS. At present, most attribute reduction methods for ordinal classification tasks mainly focus on the dependence of decision to attributes, while ignoring the information provided by the separability of classes and the correlation of attributes for ordinal classification. In view of these two issues, this article first proposes a robust FDRS model with filterable noise samples. Then, the class-separability and attribute-correlation are explored in robust fuzzy dominance rough approximation space, and corresponding attribute evaluation index is designed. Finally, an attribute reduction algorithm is designed to select the attribute subset with the highest classification performance. The experimental results show that the proposed algorithm has better robustness and classification performance.

This paper studies the problem of continual learning in an open-world scenario, referred to as Open-world Continual Learning (OwCL). OwCL is increasingly rising while it is highly challenging in two-fold: i) learning a sequence of tasks without forgetting knowns in the past, and ii) identifying unknowns (novel objects/classes) in the future. Existing OwCL methods suffer from the adaptability of task-aware boundaries between knowns and unknowns, and do not consider the mechanism of knowledge transfer. In this work, we propose Pro-KT, a novel prompt-enhanced knowledge transfer model for OwCL. Pro-KT includes two key components: (1) a prompt bank to encode and transfer both task-generic and task-specific knowledge, and (2) a task-aware open-set boundary to identify unknowns in the new tasks. Experimental results using two real-world datasets demonstrate that the proposed Pro-KT outperforms the state-of-the-art counterparts in both the detection of unknowns and the classification of knowns markedly. Code released at https://github.com/YujieLi42/Pro-KT.

The effective management of trajectory data heavily relies on the utilization of fundamental spatio-temporal queries. The surge in trajectory data, with its dynamic spatio-temporal properties, poses notable management challenges. Existing systems are inadequate in providing fine-grained trajectory representations and efficient architecture for processing queries, leading to significant computational overhead. This paper introduces TMan to address these challenges. First, TMan presents two innovative index structures that precisely capture the spatio-temporal characteristics of trajectory data. Compared to the state-of-the-art indexes, our indexes for temporal range and spatial range queries can reduce the number of retrievals by up to 77% and 83%, respectively. Next, TMan devises concise and effective encoding methods for these indexes. Leveraging these indexes, TMan provides a distributed storage structure and an index caching mechanism for efficiently managing trajectories in key-value data stores. Moreover, TMan introduces a parallel query processing approach incorporating a push-down strategy to enhance the efficiency of fundamental queries. Extensive experimental results demonstrate that TMan's index structures and architecture outperform the baselines.

Spatio-temporal forecasting of future values of spatially correlated time series is important across many cyber-physical systems (CPS). Recent studies offer evidence that the use of graph neural networks to capture latent correlations between time series holds a potential for enhanced forecasting. However, most existing methods rely on predefined or self-learning graphs, which are either static or unintentionally dynamic, and thus cannot model the time-varying correlations that exhibit trends and periodicities caused by the regularity of the underlying processes in CPS. To tackle such limitation, we propose Time-aware Graph Structure Learning (TagSL), which extracts time-aware correlations among time series by measuring the interaction of node and time representations in high-dimensional spaces. Notably, we introduce time discrepancy learning that utilizes contrastive learning with distance-based regularization terms to constrain learned spatial correlations to a trend sequence. Additionally, we propose a periodic discriminant function to enable the capture of periodic changes from the state of nodes. Next, we present a Graph Convolution-based Gated Recurrent Unit (GCGRU) that jointly captures spatial and temporal dependencies while learning time-aware and node-specific patterns. Finally, we introduce a unified framework named Time-aware Graph Convolutional Recurrent Network (TGCRN), combining TagSL, and GCGRU in an encoder-decoder architecture for multi-step spatiotemporal forecasting. We report on experiments with TGCRN and popular existing approaches on five real-world datasets, thus providing evidence that TGCRN is capable of advancing the state-of-the-art. We also cover a detailed ablation study and visualization analysis, offering detailed insight into the effectiveness of time-aware structure learning.

Scene-Text Visual Question Answering (ST-VQA) aims to understand scene text in images and answer questions related to the text content. Most existing methods heavily rely on the accuracy of Optical Character Recognition (OCR) systems, and aggressive fine-tuning based on limited spatial location information and erroneous OCR text information often leads to inevitable overfitting. In this paper, we propose a multimodal adversarial training architecture with spatial awareness capabilities. Specifically, we introduce an Adversarial OCR Enhancement (AOE) module, which leverages adversarial training in the embedding space of OCR modality to enhance fault-tolerant representation of OCR texts, thereby reducing noise caused by OCR errors. Simultaneously, We add a Spatial-Aware Self-Attention (SASA) mechanism to help the model better capture the spatial relationships among OCR tokens. Various experiments demonstrate that our method achieves significant performance improvements on both the ST-VQA and TextVQA datasets and provides a novel paradigm for multimodal adversarial training.

Indoor temperature prediction is crucial for decision-making in central heating systems. Beyond accuracy, predictions shall be interpretable, i.e. conform to the laws of physics; otherwise, it may lead to system failures or unsafe conditions. However, deep learning models often face criticism regarding interpretability, which limits their application in such settings. To this end, we propose a Spatio-Temporal Consistency enhanced Differential Network (CONST) for interpretable indoor temperature prediction. Our approach mainly consists of a differential predictive module and a spatio-temporal consistency module. Modeling the influential factors, the first module solves the issue of multicollinearity through the differential operation. Considering the heterogeneity of global and local data distributions, the second module characterizes the temporal and spatial consistency to mine the universal pattern by multi-task learning, thereby improving the prediction interpretability. Besides, we propose a set of interpretability metrics to overcome the drawbacks of partial dependence plot metric, which are more practical, zero-centered, flexible, and numerical. We conclude experiments on a real-world dataset with four heating stations. The results demonstrate the advantages of our approach over various baselines, where the interpretability can be improved by more than 8 times on cRPD while maintaining high accuracy. We developed CONST on the SmartHeat system, providing hourly indoor temperature forecasts for 13 heating stations in northern China.

Personalized Federated Continual Learning (PFCL) is a new practical scenario that poses greater challenges in sharing and personalizing knowledge. PFCL not only relies on knowledge fusion for server aggregation at the global spatial-temporal perspective but also needs model improvement for each client according to the local requirements. Existing methods, whether in Personalized Federated Learning (PFL) or Federated Continual Learning (FCL), have overlooked the multi-granularity representation of knowledge, which can be utilized to overcome Spatial-Temporal Catastrophic Forgetting (STCF) and adopt generalized knowledge to itself by coarse-to-fine human cognitive mechanisms. Moreover, it allows more effectively to personalized shared knowledge, thus serving its own purpose. To this end, we propose a novel concept called multi-granularity prompt, i.e., coarse-grained global prompt acquired through the common model learning process, and fine-grained local prompt used to personalize the generalized representation. The former focuses on efficiently transferring shared global knowledge without spatial forgetting, and the latter emphasizes specific learning of personalized local knowledge to overcome temporal forgetting. In addition, we design a selective prompt fusion mechanism for aggregating knowledge of global prompts distilled from different clients. By the exclusive fusion of coarse-grained knowledge, we achieve the transmission and refinement of common knowledge among clients, further enhancing the performance of personalization. Extensive experiments demonstrate the effectiveness of the proposed method in addressing STCF as well as improving personalized performance.

Large pre-trained vision-language models like CLIP have shown amazing zero-shot recognition performance. To adapt pre-trained vision-language models to downstream tasks, recent studies have focused on the learnable context + class name paradigm, which learns continuous prompt contexts on downstream datasets. In practice, the learned prompt context tends to overfit the base categories and cannot generalize well to novel categories out of the training data. Recent works have also noticed this problem and have proposed several improvements. In this work, we draw a new insight based on empirical analysis, that is, uninformative class names lead to degraded base-to-novel generalization performance in prompt learning, which is usually overlooked by existing works. Under this motivation, we advocate to improve the base-to-novel generalization performance of prompt learning by enhancing the semantic richness of class names. We coin our approach as the Information Disengagement based Associative Prompt Learning (IDAPL) mechanism which considers the associative, meanwhile, decoupled learning of prompt context and class name embedding. IDAPL can effectively alleviate the phenomenon of learnable context overfitting to base classes, meanwhile, learning more informative semantic representation of base classes by fine-tuning the class name embedding, leading to improved performance on both base and novel classes. Experimental results on eleven widely used few-shot learning benchmarks clearly validate the effectiveness of our proposed approach. Code is available at https://github.com/tiggers23/IDAPL

This paper delves into federated class-incremental learning (FCiL), where new classes appear continually or even privately to local clients. However, existing FCiL methods suffer from the problem of spatial-temporal catastrophic forgetting, i.e., forgetting the previously learned knowledge over time and the client-specific information owned by different clients. Additionally, private class and knowledge heterogeneity amongst local clients further exacerbate spatial-temporal forgetting, making FCiL challenging to apply. To address these issues, we propose Federated Class-specific Binary Classifier (FedCBC), an innovative approach to transferring and fusing knowledge across both temporal and spatial perspectives. FedCBC consists of two novel components: (1) continual personalization that distills previous knowledge from a global model to multiple local models, and (2) selective knowledge fusion that enhances knowledge integration of the same class from divergent clients and shares private knowledge with other clients. Extensive experiments using three newly-formulated metrics (termed GA, KRS, and KRT) demonstrate the effectiveness of the proposed approach. Our code is now hosted at: https://github.com/SkyOfBeginning/FedCBC.

Transformer-based deep models for single image super-resolution (SISR) have greatly improved the performance of lightweight SISR tasks in recent years. However, they often suffer from heavy computational burden and slow inference due to the complex calculation of multi-head self-attention (MSA), seriously hindering their practical application and deployment. In this work, we present an efficient SR model to mitigate the dilemma between model efficiency and SR performance, which is dubbed Entropy Attention and Receptive Field Augmentation network (EARFA), and composed of a novel entropy attention (EA) and a shifting large kernel attention (SLKA). From the perspective of information theory, EA increases the entropy of intermediate features conditioned on a Gaussian distribution, providing more informative input for subsequent reasoning. On the other hand, SLKA extends the receptive field of SR models with the assistance of channel shifting, which also favors to boost the diversity of hierarchical features. Since the implementation of EA and SLKA does not involve complex computations (such as extensive matrix multiplications), the proposed method can achieve faster nonlinear inference than Transformer-based SR models while maintaining better SR performance. Extensive experiments show that the proposed model can significantly reduce the delay of model inference while achieving the SR performance comparable with other advanced models.

Abstract With the exponential increase in the urban population, urban transportation systems are confronted with numerous challenges. Traffic congestion is common, traffic accidents happen frequently, and traffic environments are deteriorating. To alleviate these issues and improve the efficiency of urban transportation, accurate traffic forecasting is crucial. In this study, we aim to provide a comprehensive overview of the overall architecture of traffic forecasting, covering aspects such as traffic data analysis, traffic data modeling, and traffic forecasting applications. We begin by introducing existing traffic forecasting surveys and preliminaries. Next, we delve into traffic data analysis from traffic data collection, traffic data formats, and traffic data characteristics. Additionally, we summarize traffic data modeling from spatial representation, temporal representation, and spatio-temporal representation. Furthermore, we discuss the application of traffic forecasting, including traffic flow forecasting, traffic speed forecasting, traffic demand forecasting, and other hybrid traffic forecasting. To support future research in this field, we also provide information on open datasets, source resources, challenges, and potential research directions. As far as we know, this paper represents the first comprehensive survey that focuses specifically on the overall architecture of traffic forecasting.

In this paper, we propose a novel self-supervised Gaussian Restricted Boltzmann Machine with contrastive learning (CL-GRBM), which fuses contrastive representation learning and contrastive divergence to optimize and enhance the representation of GRBM. Built upon the concept of contrastive representation learning, CL-GRBM aims to enhance the representation capacity of the GRBM. Firstly, a pair of positive samples are constructed by one times Gibbs sampling with the original data. Then, the contrastive loss is used to pull the positive samples closer and push other samples further apart, making similar representations closer and different representations farther apart. In summary, during the training process of CL-GRBM using the CD algorithm, the objective is to align the sampling probability distribution of the visible layer in CL-GRBM as closely as possible with the empirical distribution of the original data. In the feature space of the sampling probabilities in the hidden layer, the distance between positive samples is minimized to capture the intrinsic structure of the data. According to the experimental verification, the proposed CL-GRBM shows better performance than contrastive models. As a shallow self-supervised model, it has even better performance than some excellent deep self-supervised models.

Deploying models on target domain data subject to distribution shift requires adaptation. Test-time training (TTT) emerges as a solution to this adaptation under a realistic scenario where access to full source domain data is not available, and instant inference on the target domain is required. Despite many efforts into TTT, there is a confusion over the experimental settings, thus leading to unfair comparisons. In this work, we first revisit TTT assumptions and categorize TTT protocols by two key factors, i.e., whether testing data is sequentially streamed and whether source model is allowed to be trained with modified loss function. Among the multiple protocols, we adopt a realistic sequential test-time training (sTTT) protocol, under which we develop a test-time anchored clustering (TTAC) approach to enable stronger test-time feature learning. TTAC discovers clusters in both source and target domains and matches the target clusters to the source ones to improve adaptation. When source domain information is strictly absent (i.e., source-free) we further develop an efficient method to infer source domain distributions for anchored clustering. Finally, self-training (ST) has demonstrated great success in learning from unlabeled data and we empirically figure out that applying ST alone to TTT is prone to confirmation bias. Therefore, a more effective TTT approach is introduced by regularizing self-training with anchored clustering, and the improved model is referred to as TTAC++. We demonstrate that, under all TTT protocols, TTAC++ consistently outperforms the state-of-the-art methods on five TTT datasets, including corrupted target domain, selected hard samples, synthetic-to-real adaptation and adversarially attacked target domain. We hope this work will provide a fair benchmarking of TTT methods, and future research should be compared within respective protocols.

Feature selection is a crucial step for data preprocessing, and it is widely applied in machine learning. It can eliminate features that are redundant or irrelevant from data, thereby improving performance and reducing runtime. The uncertain nature of labels produces unique challenges for high-dimensional data with label ambiguity, which is still an open problem; the structural information of the data is not utilized fully. In this article, we sufficiently consider the structural information of the data, including relevancy between labels and features, redundancy among features, and positive regions, and set up a novel label ambiguity feature selection model via weighted label-fuzzy relevancy and redundancy. Specifically, we first transform the non-label distribution annotations to label distribution annotations by using a label enhancement model. Second, we use a fuzzy similarity relation to quantify how similar samples are in label space. Third, a general label-fuzzy rough set model is created, and then, a novel feature evaluation measure based on weighted label-fuzzy relevancy and redundancy is defined. In this model, general label-fuzzy rough sets are employed to process label ambiguity problems, and the label-fuzzy relevancy and redundancy are weighted with the feature significance with the positive region as the focus. Finally, a feature selection algorithm for label ambiguity that follows the idea of weighted label-fuzzy relevancy and redundancy is proposed. Extensive experiments are conducted on 12 label distribution annotation datasets and 8 multi-label annotation datasets. The results indicate the advantages of our proposed algorithm over state-of-the-art algorithms.

Many models and methods have been developed to determine numerical thresholds in three-way decision. However, there are challenges such as local optimum-based thresholds, initial point-related optimal solutions, and a limited model generalization. To address these challenges, we design a penalty mechanism-based approximate model and solving algorithm. First, a standard optimization model and its approximate model are established by means of a penalty mechanism. Moreover, three properties of the approximate model are explored, and the relationships between both types of models are analyzed. Second, a penalty mechanism-based particle swarm optimization (PMPSO) algorithm is designed to solve general frameworks with our established models, and comparative experiments are conducted to verify the effectiveness and advantages of the algorithm. Third, we generalize the established model by fuzzy loss representations, and take linguistic intuitionistic fuzzy numbers (LIFNs) as a representation to establish general threshold-determined models based on single and multiple ranking measure functions of LIFNs. Then, we prove the existence and uniqueness of the optimal solution and develop a three-way decision method with the PMPSO algorithm. Finally, an illustrative example and more comparative analyses are considered to demonstrate the effectiveness of our method.

Feature extraction is a fundamental and challenging task in machine learning, which aims at extracting a subset of significant and discriminant features from raw data for various downstream tasks. The extraction process involves mapping the original data into a space with lower dimensions while preserving the desirable information. However, the data often has hidden manifold structures, which contain neighbor sample information within the same class. Most existing methods that extract data features without considering the potentially significant manifold structures would result in poorly discriminative features. To address these challenges, we propose a novel geodesic fuzzy rough set model (GFRS) to capture those complex manifold structures embed in the data. Given GFRS, we further design a discriminant feature extraction algorithm based on graph embedding to enhance the discriminative performance of the extracted features. Extensive experiments on fourteen real-world datasets and visualization results on modified national institute of standards and technology (MNIST) digits demonstrate the effectiveness of the proposed algorithm and its superiority over baselines methods.

Co-clustering is the simultaneous clustering of the samples and attributes of a data matrix that provides deeper insight into data than traditional clustering. However, there is a lack of representation learning algorithms that serve this mechanism of co-clustering, and the current representation learning algorithms are limited to the sample perspective and lack the use of information in the attribute perspective. To solve this problem, in this article, ctSNE , a co-representation learning model based on t-distributed stochastic neighbor embedding, is proposed for unsupervised co-clustering, where ctSNE makes the dataset representation outputted more discriminative of row and column clusters (i.e. co-discrimination). On the basis of t-distributed stochastic neighbor embedding retaining the sample data distribution and local data structure, the philosophy of collaboration is introduced (i.e., row and column hidden relationship information) so that the ctSNE model is equipped with co-representation learning capability, which can effectively improve the performance of co-clustering. To prove the effectiveness of the ctSNE model, several classic co-clustering algorithms are used to check the co-representation performance of ctSNE, and a novel internal index based on an internal clustering index, known as total inertia, is proposed to demonstrate the effect of co-clustering. The numerous experimental results show that ctSNE has tremendous co-representation capability and can significantly improve the performance of co-clustering algorithms.

Representation learning is an important topic in machine learning, pattern recognition, and data mining research. Among many representation learning approaches, semi-nonnegative matrix factorization (SNMF) is a frequently-used one. However, a typical problem of SNMF is that usually there is no learning rate guidance during the optimization process, which often leads to a poor representation ability. To overcome this limitation, we propose a very general representation learning framework (DNSRF) that is based on a deep neural net. Essentially, the parameters of the deep net used to construct the DNSRF algorithms are obtained by matrix element update. In combination with different activation functions, DNSRF can be implemented in various ways. In our experiments, we tested nine instances of our DNSRF framework on six benchmark datasets. In comparison with other state-of-the-art methods, the results demonstrate the superior performance of our framework, which is thus shown to have a great representation ability.

3D object detection has recently received much attention due to its great potential in autonomous vehicle (AV). The success of deep learning based object detectors relies on the availability of large-scale annotated datasets, which is time-consuming and expensive to compile, especially for 3D bounding box annotation. In this work, we investigate diversity-based active learning (AL) as a potential solution to alleviate the annotation burden. Given limited annotation budget, only the most informative frames and objects are automatically selected for human to annotate. Technically, we take the advantage of the multimodal information provided in an AV dataset, and propose a novel acquisition function that enforces spatial and temporal diversity in the selected samples. We benchmark the proposed method against other AL strategies under realistic annotation cost measurements, where the realistic costs for annotating a frame and a 3D bounding box are both taken into consideration. We demonstrate the effectiveness of the proposed method on the nuScenes dataset and show that it outperforms existing AL strategies significantly.

The rapid growth of Internet services dramatically drives the development of various intelligent technologies. As an important composition, modern ride-hailing platforms allow citizens to order taxis in a fast, simple, and secure manner. The key to developing a successful ride-hailing system largely depends on efficient fleet management that narrows the demand-supply gap. For this reason, many Deep Reinforcement Learning (DRL) frameworks are becoming increasingly popular for proactive taxi dispatching. However, existing approaches have very coarse formulations and solutions to the dispatching problem, which overly simplify the real situations. To this end, we propose a fine-grained DRL framework for effective taxi dispatching with tractable solutions. Our framework features three components 1) The basic model is a multi-agent DRL that optimizes the standard objectives (i.e., fulfilling the taxi orders). To reduce the status space and environment dynamics, we cluster regions based on their daily moving patterns and allow agents in the same cluster to learn cooperatively by sharing their policies, i.e., learning from like-pattern peers. 2) We further employ a recurrent neural network to forecast the demand and supply, which helps the agents make globally optimal decisions 3) Finally, we design a two-phase central dispatching strategy based on Maximum Network Flow to relocate taxis in fine granularity. We conduct extensive experiments in a realistic environment simulator using a real-world dataset, and the results demonstrate the superior performance of our new framework over existing approaches, promoting the average total order response rate by$2.67\%$, taxi effort gain by$25.77\%$, and decreasing the average number of repositions by$2.96\%$.

Co-clustering is to cluster samples and features simultaneously, which can also reveal the relationship between row clusters and column clusters. Therefore, lots of scientists have drawn much attention to conduct extensive research on it, and co-clustering is widely used in recommendation systems, gene analysis, medical data analysis, natural language processing, image analysis, and social network analysis. In this article, we survey the entire research aspect of co-clustering, especially the latest advances in co-clustering, and discover the current research challenges and future directions. First, due to different views from researchers on the definition of co-clustering, this article summarizes the definition of co-clustering and its extended definitions, as well as related issues, based on the perspectives of various scientists. Second, existing co-clustering techniques are approximately categorized into four classes: information-theory-based, graph-theory-based, matrix-factorization-based, and other theories-based. Third, co-clustering is applied in various aspects such as recommendation systems, medical data analysis, natural language processing, image analysis, and social network analysis. Furthermore, 10 popular co-clustering algorithms are empirically studied on 10 benchmark datasets with 4 metrics—accuracy, purity, block discriminant index, and running time, and their results are objectively reported. Finally, future work is provided to get insights into the research challenges of co-clustering.

Competitor identification is an essential component of corporate strategy. With the rapid development of artificial intelligence, various data-mining methodologies and frameworks have emerged to identify competitors. In general, the competitiveness among companies is determined by both market commonality and resource similarity. However, because resource information is more difficult to obtain than market information, existing studies primarily identify competitors via market commonality. To address this limitation, we introduce multisource company descriptions as well as heterogeneous business relationships, and we propose a novel method for simultaneously mining the market commonality and resource similarity. First, we use multisource company descriptions to represent companies and transform the heterogeneous business relationships into a heterogeneous business network. Then, we propose a novel multisource data fusion-based heterogeneous graph attention network (MHGAT) to learn the pairwise competitive relationships between companies. Specifically, a graph neural network-based model is proposed to learn the embeddings of companies by preserving their competition, and a multilevel attention framework is designed to integrate the embeddings from neighboring company level, heterogeneous relationship level, and multisource description level. Finally, experiments on a real-world dataset verify the effectiveness of our proposed MHGAT and demonstrate the usefulness of company descriptions and business relationships in competitor identification.

This paper presents a novel framework for continual feature selection (CFS) in data preprocessing, particularly in the context of an open and dynamic environment where unknown classes may emerge. CFS encounters two primary challenges: the discovery of unknown knowledge and the transfer of known knowledge. To this end, we propose a GBCFS method, which combines the strengths of continual learning (CL) with granular-ball computing (GBC). The GBCFS method focuses on constructing a granular-ball knowledge base to detect unknown classes and facilitate the transfer of previously learned knowledge for further feature selection. GBCFS consists of two stages: initial learning and open learning. The former aims to establish an initial knowledge base through multi-granularity representation using granular balls. The latter utilizes prior granular-ball knowledge to identify unknowns, updates the knowledge base for granular-ball knowledge transfer, reinforces old knowledge, and integrates new knowledge. Subsequently, we devise an optimal feature subset mechanism that incorporates minimal new features into the existing optimal subset, often yielding superior results during each period. Extensive experimental results on public benchmark datasets demonstrate our method's superiority in terms of both effectiveness and efficiency compared to state-of-the-art feature selection methods.

Fraud detection poses a fundamental yet challenging problem to mitigate various risks associated with fraudulent activities. However, existing methods are limited by their reliance on static data within single geographical regions, thereby restricting the trained model’s adaptability across different regions. Practically, when enterprises expand their business into new cities or countries, training a new model from scratch can incur high computational costs and lead to catastrophic forgetting (CF). To address these limitations, we propose cross-regional fraud detection as an incremental learning problem, enabling the development of a unified model capable of adapting across diverse regions without suffering from CF. Subsequently, we introduce Cross-Regional Continual Learning (CCL), a novel paradigm that facilitates knowledge transfer and maintains performance when incrementally training models from previously learned regions to new ones. Specifically, CCL utilizes prototype-based knowledge replay for effective knowledge transfer while implementing a parameter smoothing mechanism to alleviate forgetting. Furthermore, we construct heterogeneous trade graphs (HTGs) and leverage graph-based backbones to enhance knowledge representation and facilitate knowledge transfer by uncovering intricate semantics inherent in cross-regional datasets. Extensive experiments demonstrate the superiority of our proposed method over baseline approaches and its substantial improvement in cross-regional fraud detection performance.

Learning classification models from real-world data often requires substantial human effort devoted to instance annotation. As the instance-based annotating process can be very time-consuming and costly, we propose a novel active learning framework that builds classification models from human-annotatedregions. A region is defined by a set of conjunctive patterns that are formed by value ranges over the input features. A region label is a human assessment of the classproportionin the data population covered by the region. By leveraginglearning from label proportionsalgorithms, regions and their class proportions can be used to train instance-based classification models. However, the key challenge is that in practice, very few regions are defined already. Therefore, to identify regions important for model learning, we design ahierarchical active learning(HAL) framework, which actively builds a hierarchy of regions. Similar to the decision-tree learning process, our approach progressively divides the input data space into smaller sub-regions, solicits labels for the new regions, and retrains the base classification model with all the leaf regions. And we further develop amulti-hierarchy(forest) solution, which builds multiple shallower hierarchies that have more informative, diverse, and simpler regions. We evaluate our HAL framework on numerous impactful classification datasets as well as on a real user study - on the survival analysis of colorectal cancer patients. The results demonstrate that region-based active learning methods can learn high-quality classifiers from very few labeled regions. Hence, our framework is shown very effective in reducing the human annotation effort needed for building classification models.

With the digitization of modern cities, large data volumes and powerful computational resources facilitate the rapid update of intelligent models deployed in smart cities. Continual learning (CL) is a novel machine learning paradigm that constantly updates models to adapt to changing environments, where the learning tasks, data, and distributions can vary over time. Our survey provides a comprehensive review of continual learning methods that are widely used in smart city development. The content consists of three parts: 1) Methodology-wise. We categorize a large number of basic CL methods and advanced CL frameworks in combination with other learning paradigms including graph learning, spatial-temporal learning, multi-modal learning, and federated learning. 2) Application-wise. We present numerous CL applications covering transportation, environment, public health, safety, networks, and associated datasets related to urban computing. 3) Challenges. We discuss current problems and challenges and envision several promising research directions. We believe this survey can help relevant researchers quickly familiarize themselves with the current state of continual learning research used in smart city development and direct them to future research trends.

Data privacy and silos are nontrivial and greatly challenging in many real-world applications. Federated learning is a decentralized approach to training models across multiple local clients without the exchange of raw data from client devices to global servers. However, existing works focus on a static data environment and ignore continual learning from streaming data with incremental tasks. Federated Continual Learning (FCL) is an emerging paradigm to address model learning in both federated and continual learning environments. The key objective of FCL is to fuse heterogeneous knowledge from different clients and retain knowledge of previous tasks while learning on new ones. In this work, we delineate federated learning and continual learning first and then discuss their integration, i.e., FCL, and particular FCL via knowledge fusion. In summary, our motivations are four-fold: we (1) raise a fundamental problem called “spatial-temporal catastrophic forgetting” and evaluate its impact on the performance using a well-known method called federated averaging (FedAvg), (2) integrate most of the existing FCL methods into two generic frameworks, namely synchronous FCL and asynchronous FCL, (3) categorize a large number of methods according to the mechanism involved in knowledge fusion, and finally (4) showcase an outlook on the future work of FCL.

Feature selection aims to remove irrelevant or redundant features and thereby remain relevant or informative features so that it is often preferred for alleviating the dimensionality curse, enhancing learning performance, providing better readability and interpretability, and so on. Data that contain numerical and categorical representations are called heterogeneous data, and they exist widely in many real-world applications. Neighborhood rough set (NRS) can effectively deal with heterogeneous data by using neighborhood binary relation, which has been successfully applied to heterogeneous feature selection. In this article, the NRS model as a unified framework is used to design a feature selection method to handle categorical, numerical, and heterogeneous data. First, the concept of neighborhood combination entropy (NCE) is presented. It can reflect the probability of pairs of the neighborhood granules that are probably distinguishable from each other. Then, the conditional neighborhood combination entropy (cNCE) based on NCE is proposed under the condition of considering decision attributes. Moreover, some properties and relationships between cNCE and NCE are derived. Finally, the functions of inner and outer significances are constructed to design a feature selection algorithm based on cNCE (FScNCE). The experimental results show the effectiveness and superiority of the proposed algorithm.

The expected improvement (EI) criterion has been widely used in Kriging-assisted evolutionary algorithms to select individuals for expensive evaluations. It measures the amount of improvement the candidates are expected to gain compared with the current best solution, based on which the best individuals will be picked for expensive evaluations. Since all the candidate individuals are measured by the current best solution when calculating the EI values, the population moves gradually towards the current best solution, which will decrease the diversity of the population. In this work, we propose a new anisotropic expected improvement (AEI) criterion to resolve this issue. Instead of comparing all the individuals with the current best solution, the proposed AEI compares the individuals with their corresponding parent solutions. By measuring the candidate individuals with different solutions, the AEI is able to bring more directions for the population to evolve, thus increase the diversity of the population. Numerical experiments show that the proposed AEI criterion performs significantly better than four state-of-the-art Kriging-based infill criteria. This work provides a new promising criterion for Kriging-assisted evolutionary computation.

Although deep learning has significantly facilitated the progress of single image super-resolution (SISR) in recent years, it still hits bottlenecks to further improve SR performance with the continuous growth of model scale. Therefore, one of the hotspots in the field is to construct efficient SISR models by elevating the effectiveness of feature representation. In this work, we present a straightforward and generic approach for feature enhancement that can effectively promote the performance of SR models, dubbed partial channel shifting (PCS). Specifically, it is inspired by the temporal shifting in video understanding and displaces part of the channels along the spatial dimensions, thus allowing the effective receptive field to be amplified and the feature diversity to be augmented at almost zero cost. Also, it can be assembled into off-the-shelf models as a plug-and-play component for performance boosting without extra network parameters and computational overhead. However, regulating the features with PCS encounters some issues, like shifting directions and amplitudes, proportions, patterns of shifted channels, etc. We impose some technical constraints on the issues to simplify the general channel shifting. Extensive and throughout experiments illustrate that the PCS indeed enlarges the effective receptive field, augments the feature diversity for efficiently enhancing SR recovery, and can endow obvious performance gains to existing models.

Feature learning is a crucial phase machine learning [1] – [3] . How to obtain appropriate features distribution without any background is still a hard problem in machine learning. In this paper, we present a novel unsupervised feature learning architecture (see Fig. 1 ), which consists of a multi-clustering integration module and a variant of RBM termed multi-clustering integration RBM (MIRBM). In the multi-clustering integration module, we choose three clusterers to obtain three different global clustering partitions (CPs). Then, an unanimous voting strategy is used to generate the local clustering partition (LCP) of visible layer data. Hence, the LCP only has partial visible layer data. The novel MIRBM model is a core feature encoding part of the proposed unsupervised feature learning architecture. The novelty of it is that the LCP as an unsupervised guidance is integrated into the CD 1 learning to guide the distribution of the hidden layer features. For the instance in the same LCP cluster, the hidden and reconstructed hidden layer features of the MIRBM model in the proposed architecture tend to constrict together in the training process. Meanwhile, each LCP center tends to disperse from each other as much as possible in the hidden and reconstructed hidden layer during training. This work has three main contributions: 1) A novel unsupervised feature learning architecture is proposed, which consists of a multi-clustering integration module and an MIRBM model. 2) In the multi-clustering integration module of the proposed architecture, three unsupervised algorithms are employed to obtain three different global CPs without any background knowledge or label. 3) The MIRBM model in the proposed architecture uses the LCP as an unsupervised guidance to guide the distribution of the hidden layer features by integrating the LCP into the CD 1 learning.

Recently, the contrastive language-image pre-training, e.g., CLIP, has demonstrated promising results on various downstream tasks. The pre-trained model can capture enriched visual concepts for images by learning from a large scale of text-image data. However, transferring the learned visual knowledge to open-vocabulary semantic segmentation is still under-explored. In this paper, we propose a CLIP-based model named SegCLIP for the topic of open-vocabulary segmentation in an annotation-free manner. The SegCLIP achieves segmentation based on ViT and the main idea is to gather patches with learnable centers to semantic regions through training on text-image pairs. The gathering operation can dynamically capture the semantic groups, which can be used to generate the final segmentation results. We further propose a reconstruction loss on masked patches and a superpixel-based KL loss with pseudo-labels to enhance the visual representation. Experimental results show that our model achieves comparable or superior segmentation accuracy on the PASCAL VOC 2012 (+0.3% mIoU), PASCAL Context (+2.3% mIoU), and COCO (+2.2% mIoU) compared with baselines. We release the code at https://github.com/ArrowLuo/SegCLIP.

Accurate bus ridership forecasts can help city managers develop more effective transportation plans, such as bus schedules. With the continuous development of intelligent transportation systems, many methods have been proposed to predict bus ridership. However, there are two limitations to these existing efforts. First, the existing methods mainly consider temporal properties such as closeness, period, trends, holidays, and weekends, respectively. However, the temporal characteristics can be correlated and affect each other among different time scales. Besides, the spatial interactions are dynamic and complex. Thus, how to extract and represent the high-level temporal features and the dynamic spatial dependency among multiple time scales is significant and profound for bus ridership prediction. In this paper, we propose a novel dynamic spatio-temporal multi-scale representation method DSTMR to predict bus ridership. Specifically, DSTMR consists of several dynamic spatio-temporal representation blocks (DSTs). In DST, the graph generator is used to learn the adjacency of bus stations, the graph convolution module is employed to represent the dynamic spatial relationship, and the temporal convolution module is designed to extract the high-level temporal features. To demonstrate the effectiveness of our proposed method, we first construct a dataset (ASBus) based on real-world data, which is available for further study. Extensive experiments on two real-world datasets validate the effectiveness of our method.

Stay area detection is one of the most important applications in trajectory data mining, which is helpful to understand human's behavior intentions. Traditional stay area detection methods are based on GPS data with relatively high sampling rate. However, because of privacy issues, accessing GPS data can be difficult in most real-world applications. Fortunately, traffic surveillance cameras have been widely deployed in urban area, and it provides us a novel way of acquiring vehicles' trajectories. All the vehicles that traverse by can be recognized and recorded in a passive way. However, the trajectory data collected in this way is extremely coarse, because the surveillance cameras are only deployed in important locations, such as crossroads. This coarse trajectory introduces two challenges for the stay area detection problem, i.e., whether and where the stay event occurs. In this paper, we design a two-stage method to solve the stay area detection problem with coarse trajectories. It first detects the stay event between a surveillance camera record pair, then uses a layer-by-layer stay area identification algorithm to infer the exact stay area. Extensive experiments based on real-world data were used to evaluate the performance of the proposed framework. Results demonstrate the proposed framework SAInf achieved a 58% performance improvement compared with SOTA methods.

Feature selection, also known as attribute reduction, plays a crucial role in machine learning and data mining tasks. Rough set theory-based feature selection methods have gained popularity due to their ability to handle imprecise and inconsistent data, ease of implementation, and generation of highly interpretable results. However, these methods still suffer from high computational cost when dealing with large-scale datasets with high dimensions. To overcome this shortcoming, we propose a fast attribute reduction method based on inconsistent equivalence classes. The presented method can accelerate those attribute reduction algorithms whose importance measures used can be computed using only inconsistent equivalence classes. Our proposed method improves attribute reduction efficiency through three key aspects: 1) transforming the original dataset into an equivalently simplified version with fewer samples, 2) accelerating the computation of core attributes, and 3) expediting the forward selection process by removing redundant objects and attributes. Experimental results demonstrate the high computational efficiency of our proposed method.

The instability is shown in the existing methods of representation learning based on Euclidean distance under a broad set of conditions. Furthermore, the scarcity and high cost of labels prompt us to explore more expressive representation learning methods which depends on as few labels as possible. To address above issues, the small-perturbation ideology is firstly introduced on the representation learning model based on the representation probability distribution. The positive small-perturbation information (SPI) which only depend on two labels of each cluster is used to stimulate the representation probability distribution and then two variant models are proposed to fine-tune the expected representation distribution of Restricted Boltzmann Machine (RBM), namely, Micro-supervised Disturbance Gaussian-binary RBM (Micro-DGRBM) and Micro-supervised Disturbance RBM (Micro-DRBM) models. The Kullback-Leibler (KL) divergence of SPI is minimized in the same cluster to promote the representation probability distributions to become more similar in Contrastive Divergence (CD) learning. In contrast, the KL divergence of SPI is maximized in the different clusters to enforce the representation probability distributions to become more dissimilar in CD learning. To explore the representation learning capability under the continuous stimulation of the SPI, we present a deep Micro-supervised Disturbance Learning (Micro-DL) framework based on the Micro-DGRBM and Micro-DRBM models and compare it with a similar deep structure which has no external stimulation. Experimental results demonstrate that the proposed deep Micro-DL architecture shows better performance in comparison to the baseline method, the most related shallow models and deep frameworks for clustering.

Representing the structural relations between entities, i.e., knowledge graph embedding, which is a method to learn low-dimensional representations of knowledge, has become an increasingly prevalent research orientation in cognitive and human intelligence. It is significant to study how to interrelate, fuse and embed the knowledge graph data from different domains while considering data not shared. In this paper, we propose a model of cross-domain knowledge graph embedding in federated learning (FedCKE), in which entity/relation embedding between different domains can interact securely in the case that data is not shared. In advance of client model training, we present an inter-domain encrypted entity/relation alignment method using the encrypted sample alignment method in vertical federated learning, which can obtain entity/relation intersections between different domains without revealing any triples structure and additional entities/relations in the respective datasets. On the server, we aggregate the same entity/relation embeddings by the association in conjunction with the parameter-secure aggregation method in horizontal federated learning. Experimental results on three real datasets show that the proposed FedCKE model is able to enhance the embedding of different clients (domains).

Multi-view clustering research is a hot topic in the field of data mining, where complementary information between views can better describe data objects and improve the clustering performance. Non-negative matrix factorization (NMF) based multi-view clustering algorithm suffers from weak feature extraction, slow convergence speed and low accuracy. To solve these problems, this paper proposes a generalized deep learning multi-view clustering (GDLMC) algorithm based on NMF. Firstly, via decoupling the elements in the matrix, the matrix elements are non-negatively restricted using an activation function with a non-negative value domain, and the elements are updated employing stochastic gradient descent with learning rate guidance. Then, the corresponding gradients when the elements update are transformed into generalized weights and generalized biases, followed by combining the generalized weights and generalized biases with activation functions to construct generalized deep learning (GDL). Further, GDL is adopted to learn the corresponding low-dimensional matrix of each view and consensus matrix for obtaining the GDLMC algorithm. In addition, the detailed reasoning of the GDLMC algorithm are given. Finally, extensive experiments are conducted on four public datasets including regular and large-scale datasets, and the experimental results show that GDLMC has significant advantages.

The graph regularized nonnegative matrix factorization (GNMF) algorithms have received a lot of attention in the field of machine learning and data mining, as well as the square loss method is commonly used to measure the quality of reconstructed data. However, noise is introduced when data reconstruction is performed; and the square loss method is sensitive to noise, which leads to degradation in the performance of data analysis tasks. To solve this problem, a novel graph regularized sparse NMF (GSNMF) is proposed in this article. To obtain a cleaner data matrix to approximate the high-dimensional matrix, the$l_{1}$-norm to the low-dimensional matrix is added to achieve the adjustment of data eigenvalues in the matrix and sparsity constraint. In addition, the corresponding inference and alternating iterative update algorithm to solve the optimization problem are given. Then, an extension of GSNMF, namely, graph regularized sparse nonnegative matrix trifactorization (GSNMTF), is proposed, and the detailed inference procedure is also shown. Finally, the experimental results on eight different datasets demonstrate that the proposed model has a good performance.

Feature selection has been studied by many researchers using information theory to select the most informative features. Up to now, however, little attention has been paid to the interactivity and complementarity between features and their relationships. In addition, most of the approaches do not cope well with fuzzy and uncertain data and are not adaptable to the distribution characteristics of data. Therefore, to make up for these two deficiencies, a novel interactive and complementary feature selection approach based on fuzzy multineighborhood rough set model (ICFS_FmNRS) is proposed. First, fuzzy multineighborhood granules are constructed to better adapt to the data distribution. Second, feature multicorrelations (i.e., relevancy, redundancy, interactivity, and complementarity) are considered and defined comprehensively using fuzzy multigranularity uncertainty measures. Next, the features with interactivity and complementarity are mined by the forward iterative selection strategy. Finally, compared with the benchmark approaches on several datasets, the experimental results show that ICFS_FmNRS effectively improves the classification performance of feature subsets while reducing the dimension of feature space.

Feature selection, as an effective dimensionality reduction technique, is favored in preprocessing data. However, most existing algorithms are solely liable for labeled or unlabeled data, whereas a limited portion of real-world data is annotated with labels. In this article, we therefore propose a novel scheme named SemiFREE, i.e., semisupervised feature selection with fuzzy relevance and redundancy. First, both labeled and unlabeled samples are assigned with fuzzy decisions that allow class membership to naturally express the fuzziness or uncertainty in data labeling. Second, sample similarities in feature space and fuzzy decision are captured to induce fuzzy information measures for redefining the feature relevance and redundancy. Finally, adhering to the principle of relevance-maximization and redundancy-minimization, SemiFREE leverages the forward sequential searching strategy to identify qualified features progressively. Extensive experiments demonstrate the superiority of SemiFREE in the presence of partially labeled data against some other well-established feature selection algorithms.

Feature selection methods with antinoise performance are effective dimensionality reduction methods for classification tasks with noise. However, there are few studies on robust feature selection methods for monotonic classification tasks. The fuzzy dominance rough set (FDRS) model is a nontrivial knowledge acquisition tool, which is widely used in feature selection of monotonic classification tasks. Nonetheless, this model has been proved in practice to be generally poorly fault-tolerance, and only one noisy sample can cause huge interference in acquiring knowledge. In view of these two issues, this article first designs an adaptive$K$-nearest neighbors strategy to calculate the density of samples. The noisy samples are identified according to their densities, and then an active antinoise FDRS model is proposed. Then, in the active antinoise fuzzy dominance rough approximation space, the class-separability is evaluated by the approximation operators of the proposed model, and the feature-redundancy is evaluated by the fuzzy ranking conditional mutual information. On this basis, a feature evaluation index is designed comprehensively considering class-separability and feature-redundancy. Finally, a feature selection algorithm is designed to select the feature subset with the highest classification performance. The experimental results show that the proposed algorithm has better robustness and classification performance.

Most extant feature selection works neglect interactive features in the form of groups, leading to the omission of some important discriminative information. Moreover, the prevalence of data with uncertainty, fuzziness, and noise poses a certain obstacle to feature selection. Driven by these two issues, a Feature Grouping and Selection approach in Robust Fuzzy Rough Approximation Space using graph theory (FGS-RFRAS) is proposed in this study. First, a robust fuzzy rough approximation space is constructed by a neighborhood adaptive$\beta$-precision fuzzy rough set model to enhance the robustness and antinoise ability of the fuzzy rough set model. Second, uncertainty measures in robust fuzzy rough approximation space are defined to analyze the interactivity and redundancy of pairwise features on graph structure. Then, a strategy ofInteractive Retainment, Weakly Correlated Removal, and Max-Dependent Selectionis devised to guide feature grouping and selection. Experiments are performed on 21 datasets to evaluate the performance of FGS-RFRAS and demonstrate its significance. The robustness test indicates that it is antinoise for mislabeling.

In practice, it is common that there will be the same decision results under different scale conditions. Therefore, knowledge representation based on a single scale feature framework is far from meeting the needs of practical applications. Based on this, multiscale data have received extensive attention. Feature selection is an important application of fuzzy multigranularity data analysis model. The existing multiscale fuzzy granulation-based feature selection methods remove redundant or irrelevant features by selecting the optimal scale. However, this will lose the information corresponding to the remaining scale fuzzy granules, which will affect the classification results or learning tasks. Inspired by this, multiscale fuzzy entropy is defined to fuse the granule information at different scales, and applied to feature selection. First, the feature with maximum multiscale fuzzy mutual information is first selected. Then, the most significant features are gradually selected by evaluation metric that simultaneously considers the redundancy, relevance, and complementary. A multiscale fuzzy entropy-based feature selection algorithm by means of this evaluation index is further designed. Finally, the proposed method is compared with some state-of-the-art methods. The experimental results show that the proposed algorithm has higher reduction efficiency than the comparison algorithms.

Fuzzy rough set theory can model uncertainty in data and has been applied to feature selection for machine learning tasks. The existence of noise in data is one of the reasons for data uncertainty. However, most classical fuzzy rough set models are often sensitive to the noise in data, which somewhat degrades their applicability to process uncertainty of data. Furthermore, a robust feature evaluation function is nontrivial in a fuzzy rough set model as a nonoptimal feature subsets may be selected due to the perturbations from redundant features. In this article, we delve into local density and indispensable features for fuzzy rough feature selection to address these challenges. We first propose a local density-based fuzzy rough set (LDFRS) model to tackle noisy data. Mutual information is then plugged into the proposed LDFRS model to evaluate uncertainty in data. A joint feature evaluation function on the indispensability and relevance of features is constructed to evaluate the significance of features. On this basis, a fuzzy rough feature selection algorithm is built upon the LDFRS model. Experimental results using four typical classifiers demonstrate the robustness and effectiveness of the proposed model including our feature selection algorithm and its superiority against baseline methods.

The performance of multilabel learning depends heavily on the quality of the input features. A mass of irrelevant and redundant features may seriously affect the performance of multilabel learning, and feature selection is an effective technique to solve this problem. However, most multilabel feature selection methods mainly emphasize removing these useless features, and the exploration of feature interaction is ignored. Moreover, the widespread existence of real-world data with uncertainty, ambiguity, and noise limits the performance of feature selection. To this end, our work is dedicated to designing an efficient and robust multilabel feature selection scheme. First, the distribution character of multilabel data is analyzed to generate robust fuzzy multineighborhood granules. By exploring the classification information implied in the data under the granularity structure, a robust multilabel$k$-nearest neighbor fuzzy rough set model is constructed, and the concept of fuzzy dependency is studied. Second, a series of fuzzy multineighborhood uncertainty measures in$k$-nearest neighbor fuzzy rough approximation spaces are studied to analyze the correlations of feature pairs, including interactivity. Third, by investigating the uncertainty measure between feature and label, between features, multilabel data is modeled as a complete weighted graph. Then, these vertices are assessed iteratively to guide the assignment of feature weights. Finally, a graph structure-based robust multilabel feature selection algorithm (GRMFS) is designed. The experiments are conducted on 15 multilabel datasets. The results verify the superior performance of GRMFS as compared with nine representative feature selection methods.

The main goal of most information quality (IQ)-based measures is to combine data provided by multiple information sources to enhance the quality of information essential for decision makers to perform their tasks. However, there is few work to fuse multisource information from the perspective of possibility distribution (PD) and use IQ as the evaluation criteria for feature selection. The PD is one of important concepts in the possibility theory, which is a generally acknowledged method for describing a kind of uncertain knowledge. In this article, we propose a novel representation model of PDs based on FMs, namely, a possibility distribution information system (PDIS). Then, several IQ measures are defined in the PDIS, including Gini entropy, compatibility, conflict, credibility, and separability degrees. In view of this, a minimal-separability-minimal-uncertainty-based unsupervised feature selection algorithm (UmSMU) is designed. The proposed UmSMU can sufficiently fuse multiple possibilistic information. Meanwhile, the selected features maintain as much information as possible while minimizing the uncertainty of information. The experimental results show that the proposed algorithm performs well, especially when it comes to selecting fewer features and improving performance.

Learning classification models in practice usually requires numerous labeled data for training. However, instance-based annotation can be inefficient for humans to perform. In this article, we propose and study a new type of human supervision that is fast to perform and useful for model learning. Instead of labeling individual instances, humans provide supervision to dataregions, which are subspaces of the input data space, representing subpopulations of data. Since labeling now is performed on a region level, 0/1 labeling becomes imprecise. Thus, we design the region label to be aqualitativeassessment of the class proportion, which coarsely preserves the labeling precision but is also easy for humans to do. To identify informative regions for labeling and learning, we further devise ahierarchical active learningprocess that recursively constructs a region hierarchy. This process is semisupervised in the sense that it is driven by both active learning strategies and human expertise, where humans can provide discriminative features. To evaluate our framework, we conducted extensive experiments on nine datasets as well as a real user study on a survival analysis of colorectal cancer patients. The results have clearly demonstrated the superiority of our region-based active learning framework against many instance-based active learning methods.

Representation learning based on autoencoders has received great concern for its potential ability to capture valuable latent information. Conventional autoencoders pursue minimal reconstruction error, but in most machine learning tasks such as classification and clustering, the discrimination of feature representation is also important. To address this limitation, an enhanced self-supervised discriminative fuzzy autoencoder (FAE) is innovatively proposed, which focuses on exploring information within data to guide the unsupervised training process and enhancing feature discrimination in a self-supervised manner. In FAE, fuzzy membership is applied to provide a means of self-supervised, which allows FAE can not only utilize AE’s outstanding representation learning capabilities but can also transform the original data into another space with improved discrimination. First, the objective function corresponding to FAE is proposed by reconstruction loss and clustering oriented loss simultaneously. Subsequently, Mini-Batch Gradient Descent is applied to infer the objective function and the detailed process is illustrated step by step. Finally, empirical studies on clustering tasks have demonstrated the superiority of FAE over the state of the art.

Clustering is a popular research topic in the field of data mining, in which the clustering method based on non-negative matrix factorization (NMF) has been widely employed. However, in the update process of NMF, there is no learning rate to guide the update as well as the update depends on the data itself, which leads to slow convergence and low clustering accuracy. To solve these problems, a generalized deep learning clustering (GDLC) algorithm based on NMF is proposed in this article. Firstly, a nonlinear constrained NMF (NNMF) algorithm is constructed to achieve sequential updates of the elements in the matrix guided by the learning rate. Then, the gradient values corresponding to the element update are transformed into generalized weights and generalized biases, by inputting the elements as well as their corresponding generalized weights and generalized biases into the nonlinear activation function to construct the GDLC algorithm. In addition, for improving the understanding of the GDLC algorithm, its detailed inference procedure and algorithm design are provided. Finally, the experimental results on eight datasets show that the GDLC algorithm has efficient performance.

Co-clustering methods make use of the correlation between samples and attributes to explore the co-occurrence structure in data. These methods have played a significant role in gene expression analysis, image segmentation, and document clustering. In bipartite graph partition-based co-clustering methods, the relationship between samples and attributes is described by constructing a diagonal symmetric bipartite graph matrix, which is clustered by the philosophy of spectral clustering. However, this not only has high time complexity but also the same number of row and column clusters. In fact, the number of categories of rows and columns often changes in the real world. To address these problems, this paper proposes a novel fast flexible bipartite graph model for the co-clustering method (FBGPC) that directly uses the original matrix to construct the bipartite graph. Then, it uses the inflation operation to partition the bipartite graph in order to learn the co-occurrence structure of the original data matrix based on the inherent relationship between bipartite graph partitioning and co-clustering. Finally, hierarchical clustering is used to obtain the clustering results according to the set relationship of the co-occurrence structure. Extensive empirical results show the effectiveness of our proposed model and verify the faster performance, generality, and flexibility of our model.

Daily schedule recommendation is an intelligent approach to recommend multiple suitable activity locations and activity sequences for users based on their needs in a day. In such a scenario, training the model using traditional methods requires centralized data collection from individual users, which may be prohibited by data protection acts, such as GDPR and CCPA. In this paper, we address the problem of daily schedule recommendation utilizing the deep reinforcement learning model in a federated learning framework (FedDSR). And curriculum learning is applied to guide the training process towards better local optimization and better generalization. For the uploaded local parameters, a similarity aggregation algorithm is proposed to improve the quality of the model. The experimental results show that the proposed FedDSR model is superior and effective to multiple baselines on two real datasetsGeolifeandChengdu. Comparing with baselines, our method not only ensures that the parties do not need to share data and thus achieve joint modeling, but also can exceed$\sim\!\! 18\%$under evaluation metricperimeterand improve$\sim\! 0.72\%$under evaluation metricADTS.

Multi-view unsupervised feature selection (MUFS) has been demonstrated as an effective technique to reduce the dimensionality of multi-view unlabeled data. The existing methods assume that all of views are complete. However, multi-view data are usually incomplete, i.e., a part of instances are presented on some views but not all views. Besides, learning the complete similarity graph, as an important promising technology in existing MUFS methods, cannot achieve due to the missing views. In this paper, we propose a complementary and consensus learning-based incomplete multi-view unsupervised feature selection method (C$^{2}$IMUFS) to address the aforementioned issues. Concretely, C$^{2}$IMUFS integrates feature selection into an extended weighted non-negative matrix factorization model equipped with adaptive learning of view-weights and a sparse$\ell _{2,p}$-norm, which can offer better adaptability and flexibility. By the sparse linear combinations of multiple similarity matrices derived from different views, a complementary learning-guided similarity matrix reconstruction model is presented to obtain the complete similarity graph in each view. Furthermore, C$^{2}$IMUFS learns a consensus clustering indicator matrix across different views and embeds it into a spectral graph term to preserve the local geometric structure. Comprehensive experimental results on real-world datasets demonstrate the effectiveness of C$^{2}$IMUFS compared with state-of-the-art methods.

Recommender system can provide users with the required information accurately and efficiently, playing a very important role in improving users' life experience. Although knowledge graph-based recommender system can solve the sparsity and cold start problems faced by traditional recommender system, it cannot handle the cross-domain cold start problem and cannot provide multi-domain recommendations. Therefore, this paper focuses on multi-domain item-item (I2I) recommendation based on cross-domain knowledge graph embedding by analyzing the association between items of the same domain and the interaction between items of diverse domains with the aid of knowledge graph that contains rich information. Firstly, a cross-domain knowledge graph chiasmal embedding approach is proposed to efficiently interact all items in multiple domains. To help achieve both homo-domain embedding and hetero-domain embedding of items, a binding rule is put forward. Secondly, a multi-domain I2I recommendation method is presented to efficiently recommend items in multiple domains, which is a recommendation method based on link prediction of knowledge graph. Finally, the proposed methods are compared and analyzed with some benchmark methods using two datasets. The experimental results show that the proposed methods achieve better link prediction results and multi-domain recommendation results.

Urban flow analysis is an essential research for smart city construction, in which urban flow pattern analysis focuses on the continuous state of urban flow. How to mine, store and reuse traffic patterns from urban multi-source heterogeneous big data is challenging. Therefore, this paper proposes a knowledge mining network for regional flow pattern to mine and store the urban flow pattern. The proposed model consists of two modules. In the first module, the features of the region and its flow pattern are extracted as the entity and relation, respectively. In the second module, POI features are modeled to enhance the embedding representation of relation and entity. Based on the translation distance method, the knowledge triplets of regional flow patterns are mined. Finally, the proposed model is compared with some benchmark methods using Chengdu Didi order and POI datasets. Experimental results show that the proposed model is effective. In addition, the knowledge triplets are visualized and some application examples are introduced.

Feature selection refers to choose an optimal non-redundant feature subset with minimal degradation of learning performance and maximal avoidance of data overfitting. The appearance of large data explosion leads to the sequential execution of algorithms are extremely time-consuming, which necessitates the scalable parallelization of algorithms by efficiently exploiting the distributed computational capabilities. In this paper, we present parallel feature selection algorithms underpinned by a rough hypercuboid approach in order to scale for the growing data volumes. Metrics in terms of rough hypercuboid are highly suitable to parallel distributed processing, and fits well with the Apache Spark cluster computing paradigm. Two data parallelism strategies, namely, vertical partitioning and horizontal partitioning, are implemented respectively to decompose the data into concurrent iterative computing streams. Experimental results on representative datasets show that our algorithms significantly faster than its original sequential counterpart while guaranteeing the quality of the results. Furthermore, the proposed algorithms are perfectly capable of exploiting the distributed-memory clusters to accomplish the computation task that fails on a single node due to the memory constraints. Parallel scalability and extensibility analysis have confirmed that our parallelization extends well to process massive amount of data and can scales well with the increase of computational nodes.

The International Classification of Diseases (ICD) is a standard for categorizing physical conditions, which has been widely used for analyzing clinical data and monitoring health issues. Manual ICD coding takes a long time and is vulnerable to errors, so people pay more and more attention to the application of deep neural networks in ICD automatic coding. However, there is still no comprehensive review of these studies and prospects for further research. This paper is not limited to the study of deep neural networks, but gives a formal definition of ICD coding problems, and then systematically reviews the existing literature on how to design deep neural networks to address the four major challenges of ICD coding tasks. This paper also summarizes the public data sets and future research directions, to provide a guidance for the research of ICD coding in medical field.

As travel efficiency matters to the work productivity of cities, shortening passengers' travel time for metros is therefore a pressing need. To this end, we study a strategy by dynamically scheduling dwell time for trains. Developing such a strategy is challenging because of three aspects: 1) Optimizing the average travel time of passengers needs to properly balance passengers' waiting time at platforms and journey time on trains, as well as considering long-term impacts; 2) Capturing dynamic spatio-temporal (ST) correlations of incoming passengers for metro stations is difficult; and 3) For each train, the dwell time scheduling is affected by other trains, which is hard to measure. To tackle these challenges, we propose a novel deep neural network, entitled AutoDwell. Specifically, AutoDwell optimizes the long-term rewards of dwell time settings in terms of passengers' waiting and journey time by a reinforcement learning framework. Next, AutoDwell employs gated recurrent units and graph attention networks to extract the ST correlations of the passenger flows among metro stations. Moreover, attention mechanisms are leveraged in AutoDwell for capturing the interactions between the trains. Extensive experiments on two real-world datasets demonstrate the superior performance of AutoDwell over several baselines, capable of saving passengers' travel time significantly.

Urban metro flow prediction is of great value for metro operation scheduling, passenger flow management and personal travel planning. However, the problem is challenging. First, different metro stations, e.g. transfer stations and non-transfer stations have unique traffic patterns. Second, it is difficult to model complex spatio-temporal dynamic relation of metro stations. To address these challenges, we develop a spatio-temporal dynamic graph relational learning model (STDGRL) to predict urban metro station flow. First, we propose a spatio-temporal node embedding representation module to capture the traffic patterns of different stations. Second, we employ a dynamic graph relationship learning module to learn dynamic spatial relationships between metro stations without a predefined graph adjacency matrix. Finally, we provide a transformer-based long-term relationship prediction module for long-term metro flow prediction. Extensive experiments are conducted based on metro data in four cities, China, with experimental results demonstrating the advantages of our method compared over 14 baselines for urban metro flow prediction.

The selection of prominent features for building more compact and efficient models is an important data preprocessing task in the field of data mining. The rough hypercuboid approach is an emerging technique that can be applied to eliminate irrelevant and redundant features, especially for the inexactness problem in approximate numerical classification. By integrating the meta-heuristic-based evolutionary search technique, a novel global search method for numerical feature selection is proposed in this article based on the hybridization of the rough hypercuboid approach and binary particle swarm optimization (BPSO) algorithm, namely RH-BPSO. To further alleviate the issue of high computational cost when processing large-scale datasets, parallelization approaches for calculating the hybrid feature evaluation criteria are presented by decomposing and recombining hypercuboid equivalence partition matrix via horizontal data partitioning. A distributed meta-heuristic optimized rough hypercuboid feature selection (DiRH-BPSO) algorithm is thus developed and embedded in the Apache Spark cloud computing model. Extensive experimental results indicate that RH-BPSO is promising and can significantly outperform the other representative feature selection algorithms in terms of classification accuracy, the cardinality of the selected feature subset, and execution efficiency. Moreover, experiments on distributed-memory multicore clusters show that DiRH-BPSO is significantly faster than its sequential counterpart and is perfectly capable of completing large-scale feature selection tasks that fail on a single node due to memory constraints. Parallel scalability and extensibility analysis also demonstrate that DiRH-BPSO could scale out and extend well with the growth of computational nodes and the volume of data.

Feature selection is an important topic in data mining and machine learning, which aims to select an optimal feature subset for building effective and explainable prediction models. This article introduces Rough Hypercuboid based Distributed Online Feature Selection (RHDOFS) method to tackle two critical challenges of Volume and Velocity associated with Big Data. By exploring the class separability in the boundary region of rough hypercuboid approach, a novel integrated feature evaluation criterion is proposed by examining not only the explicit patterns contained in the positive region but also the useful implicit patterns derived from the boundary region. An efficient online feature selection method for streaming feature scenario is developed to identify relevant and nonredundant features in an incremental iterative fashion. Furthermore, a parallel optimization mechanism by combining both data and computational independence is further employed to accelerate the original sequential implementation. An efficient distributed online feature selection algorithm is presented and implemented on the Apache Spark platform to scale for massive amount of data by exploiting the computational capabilities of multicore clusters. Encouraging results of extensive experiments indicate the superiority and notable advantages of the proposed algorithm over the relevant and representative online feature selection algorithms. Empirical tests on scalability and extensibility also demonstrate our distributed implementation significantly reduces the computational times requirements while maintaining the prediction accuracy, and is capable of scaling well in volume of data and number of computing nodes.

Fine-tuning pre-trained models for downstream tasks is mainstream in deep learning. However, the pre-trained models are limited to be fine-tuned by data from a specific modality. For example, as a visual model, DenseNet cannot directly take the textual data as its input. Hence, although the large pre-trained models such as DenseNet or BERT have a great potential for the downstream recognition tasks, they have weaknesses in leveraging multimodal information, which is a new trend of deep learning. This work focuses on fine-tuning pre-trained unimodal models with multimodal inputs of image-text pairs and expanding them for image-text multimodal recognition. To this end, we propose the Multimodal Information Injection Plug-in (MI2P) which is attached to different layers of the unimodal models (e.g., DenseNet and BERT). The proposed MI2P unit provides the path to integrate the information of other modalities into the unimodal models. Specifically, MI2P performs cross-modal feature transformation by learning the fine-grained correlations between the visual and textual features. Through the proposed MI2P unit, we can inject the language information into the vision backbone by attending the word-wise textual features to different visual channels, as well as inject the visual information into the language backbone by attending the channel-wise visual features to different textual words. Armed with the MI2P attachments, the pre-trained unimodal models can be expanded to process multimodal data without the need to change the network structures.

Similarity search has recently become an integral part of many trajectory data analysis tasks. As the number of trajectories increases, we must find similar trajectories among massive trajectories, necessitating a scalable and efficient frame-work. Typically, massive trajectory data can be managed by key-value data stores. However, existing works with key-value data stores use a coarse representation to store trajectory data. Besides, they do not provide efficient query processing to search similar trajectories. Thus, this paper proposes TraSS, an efficient framework for trajectory similarity search in key-value data stores. We propose a novel spatial index, XZ*, which utilizes fine-grained index spaces with irregular shapes and sizes to represent trajectories elaborately. Further, we devise a bijective function from the index spaces of XZ* to continuous integers, which is simple but effective for query processing. To improve the efficiency of similarity search, we employ two steps to prune dissimilar trajectories: (1) global pruning. It leverages the XZ* index to prune index spaces with no trajectories similar to the query trajectory. Our global pruning can only pick out index spaces with similar sizes and shapes to the query trajectory. Compared to the state-of-the-art index, our global pruning reduces I/O overhead up to 66.4 % during query processing; (2) local filtering. It filters dissimilar trajectories in a way with low complexity. We use a few representative features extracted from a trajectory by the Douglas-Peucker algorithm to accelerate the local filtering. We implement an open-source toolkit (TraSS) on a popular key-value data store. Extensive experiments show that TraSS outperforms state-of-the-art solutions.

Accurate traffic forecasting, the foundation of intelligent transportation systems (ITS), has never been more significant than nowadays due to the prosperity of smart cities and urban computing. Recently, Graph Neural Network truly outperforms the traditional methods. Nevertheless, the most conventional GNN-based model works well while given a predefined graph structure. And the existing methods of defining the graph structures focus purely on spatial dependencies and ignore the temporal correlation. Besides, the semantics of the static pre-defined graph adjacency applied during the whole training progress is always incomplete, thus overlooking the latent topologies that may fine-tune the model. To tackle these challenges, we propose a new traffic forecasting framework-Spatio-Temporal Latent Graph Structure Learning networks (ST-LGSL). More specifically, the model employs a graph generator based on Multilayer perceptron and K-Nearest Neighbor, which learns the latent graph topological information from the entire data considering both spatial and temporal dynamics. Furthermore, with the initialization of MLP-kNN based on ground-truth adjacency matrix and similarity metric in kNN, ST-LGSL aggregates the topologies focusing on geography and node similarity. Additionally, the generated graphs act as the input of the Spatio-temporal prediction module combined with the Diffusion Graph Convolutions and Gated Temporal Convolutions Networks. Experimental results on two benchmarking datasets in real world demonstrate that ST-LGSL outperforms various types of state-of-art baselines.

Nowadays, many cities are suffering from air pollution problems, which endangered the health of the young and elderly for breathing problems. For supporting the government’s policy-making and people’s decision making, it is important to predict future fine-grained air quality. In this article, we predict the air quality of the next 48 hours for each monitoring station and the daily average air quality of the next 7 days for a city, considering air quality data, meteorology data, and weather forecast data. Based on the domain knowledge about air pollution, we propose a deep neural network based approach, entitled DeepAir. Our approach consists of a deep distributed fusion network for station-level short-term prediction and a deep cascaded fusion network for the city-level long-term forecast. With the data transformation preprocessing, the former network adopts a neural distributed architecture to fuse heterogeneous urban data for simultaneously capturing the direct and indirect factors affecting air quality. The latter network takes a neural cascaded architecture to learn the dynamic influences from previously existing data and future predicted data on future air quality. We have deployed a real-time system on the cloud, providing fine-grained air quality forecasts for 300+ Chinese cities every hour. Our system mainly consists of three components: data crawler, task scheduler, and prediction model, which are implemented with a multi-task architecture to improve the system’s efficiency and stability. Based on the datasets from three-year nine Chinese cities, experimental results demonstrate the advantages of our proposed method.

Outlier detection is one of the most important research directions in data mining. However, most of the current research focuses on outlier detection for categorical or numerical attribute data. There are few studies on the outlier detection of mixed attribute data. In this article, we introduce fuzzy rough sets (FRSs) to deal with the problem of outlier detection in mixed attribute data. Since the outlier detection model of the classical rough set is only applicable to the categorical attribute data, we use FRS to generalize the outlier detection model and construct a generalized outlier detection model based on fuzzy rough granules. First, the granule outlier degree (GOD) is defined to characterize the outlier degree of fuzzy rough granules by employing the fuzzy approximation accuracy. Then, the outlier factor based on fuzzy rough granules is constructed by integrating the GOD and the corresponding weights to characterize the outlier degree of objects. Furthermore, the corresponding fuzzy rough granules-based outlier detection (FRGOD) algorithm is designed. The effectiveness of the FRGOD algorithm is evaluated through experiments on 16 real-world datasets. The experimental results show that the algorithm is more flexible for detecting outliers and is suitable for numerical, categorical, and mixed attribute data.

In real cases, missing values tend to contain meaningful information that should be acquired or should be analyzed before the incomplete dataset is used for machine learning tasks. In this work, two algorithms named jointly fuzzy C-Means and vaguely quantified nearest neighbor (VQNN) imputation (JFCM-VQNNI) and jointly fuzzy C-Means and fitted VQNN imputation (JFCM-FVQNNI) have been proposed by considering clustering conception and sufficient extraction of uncertain information. In the proposed JFCM-VQNNI and JFCM-FVQNNI algorithm, the missing value is regarded as a decision feature, and then, the prediction is generated for the objects that contain at least one missing value. Specially, as for JFCM-VQNNI algorithm, indistinguishable matrixes, tolerance relations, and fuzzy membership relations are adopted to identify the potential closest filled values based on corresponding similar objects and related clusters. On the basis of JFCM-VQNNI algorithm, JFCM-FVQNNI algorithm synthetic analyzes the fuzzy membership of the dependent features for instances with each cluster. In order to fill the missing values more accurately, JFCM-FVQNNI algorithm performs fuzzy decision membership adjustment in each object with respect to the related clusters by considering highly relevant decision attributes. The experiments have been carried out on five datasets. Based on the analysis of root-mean-square error, mean absolute error, comparison of imputation values with actual values, and classification accuracy results analysis, we can draw the conclusion that the proposed JFCM-FVQNNI and JFCM-VQNNI algorithms yields sufficient and reasonable imputation performance results by comparing with fuzzy C-Means parameter-based imputation algorithm and fuzzy C-Means rough parameter-based imputation algorithm.

Incremental feature selection approaches can improve the efficiency of feature selection used for dynamic datasets, which has attracted increasing research attention. Nevertheless, there is currently no work on incremental feature selection approaches for dynamic ordered data. Moreover, the monotonic classification effect of ordered data is easily affected by noise, so a robust feature evaluation metric is needed for feature selection algorithm. Motivated by these two issues, we investigate incremental feature selection approaches using a new conditional entropy with robustness for dynamic ordered data in this study. First, we propose a new rough set model, i.e., fuzzy dominance neighborhood rough sets (FDNRS). Second, a conditional entropy with robustness is defined based on FDNRS model, which is used as evaluation metric for features and combined with a heuristic feature selection algorithm. Finally, two incremental feature selection algorithms are designed on the basis of the above researches. Experiments are performed on ten public datasets to evaluate the robustness of the proposed metric and the performance of the incremental algorithms. Experimental results verify that the proposed metric is robust and our incremental algorithms are effective and efficient for updating reducts in dynamic ordered data.

Monotonic classification is a common task in the field of multicriteria decision-making, in which features and decision obey a monotonic constraint. The dominance-based rough set theory is an important mathematical tool for knowledge acquisition in monotonic classification tasks (MCTs). However, existing dominance-based rough set models are very sensitive to noise information, and only a misclassified sample will lead to large errors in acquiring knowledge. This unstable phenomenon does not meet the requirements of practical applications. On the other hand, feature selection is supposedly an effective dimensionality reduction approach for classification tasks. In the real world, feature combinations with multiple correlations can often provide important classification information, where the multiple correlations include redundancy, complementarity, and interaction between features. To the best of our knowledge, most of the existing feature selection methods for MCTs only consider the relevance between features and decision, while ignoring the multiple correlations. To overcome these two drawbacks, in this article, we propose a robust fuzzy dominance rough set model, and develop a feature selection method that considers multiple correlations based on the robust model for MCTs. First, a soft fuzzy dominance rough set (SFDRS) with robustness is proposed. Second, a feature evaluation index considering multiple correlations is presented. Finally, a feature selection algorithm based on SFDRS is designed to select an optimal feature subset. Extensive experiments are conducted on 12 public datasets, and the results show that the SFDRS model has good robustness and the proposed feature selection algorithm has excellent classification performance.

The approximation learning of fuzzy concepts associated with fuzzy rough sets and three-way decisions is a useful technology for the representation, learning, and transformation of uncertain knowledge. However, how to integrate the advantages of fuzzy rough approximations and three-way approximations has rarely been carefully investigated so far. In this article, we focus on exploring the connection and interplay of these two methods, and further propose the hierarchical fuzzy rough approximations in the dynamic fuzzy open-world environment. We utilize the temporal-spatial perspectives of three-way decisions to construct multigranularity structures and implement multigranularity learning. The temporality of data and the spatiality of model parameters are both considered in such frameworks. Subsequently, we discuss the interpretation and representation of fuzzy three-way regions in fuzzy rough sets with some definitions and properties. The max, double, and aggregated evaluation-based models of three-way approximations are proposed to gradually transform a fuzzy concept to a crisp concept by the time-evolving attributes. Finally, the comparative experimental results between static and dynamic approaches demonstrate the effectiveness of our proposed hierarchical approximation learning models.

Aiming at the problem of effectively selecting relevant features from heterogeneous data without decision, a novel feature selection approach is studied based on fuzzy mutual information in fuzzy rough set theory. First, the fuzzy relevance of each feature is defined by using fuzzy mutual information, and then, the fuzzy conditional relevance is further given. Next, the fuzzy redundancy is defined by using the difference between the fuzzy relevance and the fuzzy conditional relevance. Thereby, the evaluation index of the feature importance is obtained by using the idea of unsupervised minimum redundancy and maximum relevance. Finally, a fuzzy-mutual-information-based unsupervised feature selection algorithm is designed to select feature sequences. Extensive experiments are conducted on public datasets, and six unsupervised feature selection algorithms are compared. The selected features are evaluated by classification, clustering, and outlier detection methods. Experimental results show that the proposed algorithm can select fewer heterogeneous features to maintain or improve the performance of learning algorithms.

In this paper, we present a novel unsupervised feature learning architecture, which consists of a multi-clustering integration module and a variant of RBM termed multi-clustering integration RBM (MIRBM). In the multi-clustering integration module, we apply three clusterers (K-means, affinity propagation and spectral clustering algorithms) to obtain three different clustering partitions (CPs) without any background knowledge or label. Then, an unanimous voting strategy is used to generate a local clustering partition (LCP). The novel MIRBM model is a core feature encoding part of the proposed unsupervised feature learning architecture. The novelty of it is that the LCP as an unsupervised guidance is integrated into one step contrastive divergence (${\mathtt{{CD}}}_{1}$) learning to guide the distribution of the hidden layer features. For the instance in the same LCP cluster, the hidden and reconstructed hidden layer features of the MIRBM model in the proposed architecture tend to constrict together in the training process. Meanwhile, each LCP center tends to disperse from each other as much as possible in the hidden and reconstructed hidden layer during training. The experiments demonstrate that the proposed unsupervised feature learning architecture has more powerful feature representation and generalization capability than the state-of-the-art models for clustering tasks in the Microsoft Research Asia Multimedia (MSRA-MM)2.0 dataset.

The natural gas tightly correlates with our everyday life. However, driven by gray incomes, some users are prone to stealing gas by refitting the equipment without permission. Especially for the boiler room users in winter, this phenomenon appears more rampant. Traditional gas-theft detection methods highly rely on the on-site inspection, where exists ineffective and randomness. With the rapidly deployed IoT sensors, we can collect real-time gas consumption data to analyze users’ behavior patterns, where the gas-theft suspects could be discovered early and accurately. In this paper, we propose a data-driven approach, named SVOC, to detect gas-theft suspects among boiler room users. Our approach consists of a scenario-based data quality detection algorithm, a deformation-based normality detection algorithm, and an One-Class Support Vector Machine (OCSVM) based anomaly detection algorithm. Specifically, considering the temporal proximity between the gas consumption and the outdoor temperature, the normality detection algorithm adopts a similarity-based deformation correlation to detect normal boiler room users out of abnormal ones. Then, we employ OCSVM as the anomaly detection algorithm to capture various features across multiple data sources, aiming to distinguish gas-theft suspects from the remaining irregular users. Here, the detected normal and abnormal users are fed into the OCSVM for training and prediction, respectively, which can overcome the label scarcity problem. We conduct extensive experiments on a real-world dataset during one heating season. The results demonstrate distinct advantages of our approach over various baselines. We have developed a real-time system on the cloud, providing daily gas-theft suspects for gas companies.

Outlier detection is widely used in many fields, such as intrusion detection, credit card fraud detection, medical diagnosis, and so on. Existing outlier detection algorithms are mostly designed for dealing with numeric or categorical attributes. However, data usually exist in the form of mixed attributes in real-world applications. In this article, we propose a novel mixed attribute outlier detection method based on multigranulation relative entropy by employing the neighborhood rough set. First, the neighborhood system is constructed by optimizing the mixed distance metric and the radius of the statistical value. Second, the neighborhood entropy is introduced as an uncertainty measure of data. Furthermore, the three kinds of multigranulation relative entropy-based matrices are defined by three kinds of attribute sequences, and the multigranulation relative entropy-based outlier factor is integrated to indicate the outlier degree of every object. Based on the proposed outlier detection model, the corresponding algorithm is designed. Finally, the proposed algorithm is compared with other nine algorithms through experiments on public data. The experimental results show that the proposed technique is adaptive and effective.

With the rapid development of the Internet of Things (IoT), massive trajectories have been generated. Trajectory data is beneficial for many urban applications. This demo presents a holistic trajectory data management system based on distributed platforms, such as Spark and HBase, namely JUST-Traj. It provides a variety of indexes to efficiently support spatio-temporal queries and analyses on massive trajectories. Additionally, it provides a convenient SQL engine to execute all operations (storage, queries, analyses) through a SQL-like statement. Finally, we design a web portal for developers and demonstrate different operations in the portal.

Air quality forecasting has been regarded as the key problem of air pollution early warning and control management. In this article, we propose a novel deep learning model for air quality (mainly PM2.5) forecasting, which learns the spatial-temporal correlation features and interdependence of multivariate air quality related time series data by hybrid deep learning architecture. Due to the nonlinear and dynamic characteristics of multivariate air quality time series data, the base modules of our model include one-dimensional Convolutional Neural Networks (1D-CNNs) and Bi-directional Long Short-term Memory networks (Bi-LSTM). The former is to extract the local trend features and spatial correlation features, and the latter is to learn spatial-temporal dependencies. Then we design a jointly hybrid deep learning framework based on one-dimensional CNNs and Bi-LSTM for shared representation features learning of multivariate air quality related time series data. We conduct extensive experimental evaluations using two real-world datasets, and the results show that our model is capable of dealing with PM2.5 air pollution forecasting with satisfied accuracy.

Gas theft of restaurants is a major concern in the gas industry, which causes revenue losses for gas companies and endangers the public safety seriously. Traditional methods of gas theft detection highly rely on active human efforts that are extremely ineffective. Thanks to the gas consumption data collected by smart meters, we can devise a data-driven method to tackle this issue. In this paper, we propose a gas-theft detection method msRank to discover suspicious restaurant users when only scarce labels are available. Our method contains three main components: 1)data pre-processing, which filters reading noises and excludes data-missing or zero-use users; 2)normal user modeling, which quantifies the self-stable seasonality of normal users and distinguishes them from unstable ones; and 3)gas-theft suspect detection, which discovers gas-theft suspects among unstable users by RankNet-based suspicion scoring on extracted deviation features. By using detected normal users as negative samples to train RankNet, the component of normal user modeling and that of gas-theft suspect detection are seamlessly connected, overcoming the problem of label scarcity. We conduct extensive experiments on three real-world datasets, and the results demonstrate advantages of our approach. We have deployed a system GasShield which provides a gas-theft suspect list weekly for a gas group in northern China.

The classic deep learning paradigm learns a model from the training data of a single task and the learned model is also tested on the same task.This paper studies the problem of learning a sequence of tasks (sentiment classification tasks in our case).After each sentiment classification task is learned, its knowledge is retained to help future task learning.Following this setting, we explore attention neural networks and propose a Bayes-enhanced Lifelong Attention Network (BLAN).The key idea is to exploit the generative parameters of naïve Bayes to learn attention knowledge.The learned knowledge from each task is stored in a knowledge base and later used to build lifelong attentions.The constructed lifelong attentions are then used to enhance the attention of the network to help new task learning.Experimental results on product reviews from Amazon.com show the effectiveness of the proposed model.

Abstract Building an effective representation for 3D face geometry is essential for face analysis tasks, that is, landmark detection, face recognition and reconstruction. This paper proposes to use a Multi‐Channel Facial Shape (MCFS) representation that consists of depth, hand‐engineered feature and attention maps to construct a 3D facial descriptor. And, a multi‐channel adjustment mechanism, named filtered squeeze and reversed excitation (FSRE), is proposed to re‐organize MCFS data. To assign a suitable weight for each channel, FSRE is able to learn the importance of each layer automatically in the training phase. MCFS and FSRE blocks collaborate together effectively to build a robust 3D facial shape representation, which has an excellent discriminative ability. Extensive experimental results, testing on both high‐resolution and low‐resolution face datasets, show that facial features extracted by our framework outperform existing methods. This representation is stable against occlusions, data corruptions, expressions and pose variations. Also, unlike traditional methods of 3D face feature extraction, which always take minutes to create 3D features, our system can run in real time.

To effectively and efficiently deal with large-scale spatial data is critical for applications in the age of information technology. Compact representation of spatial knowledge is one of the emerging research techniques that contribute to this capability. In this article, we consider the problem of compactly representing qualitative directional relations between extended objects, modelled in the Cardinal Direction Calculus (CDC) of Goyal and Egenhofer. For a large dataset of regions, this approach first constructs a simplified geometry for each region, which preserves CDC relations between regions, and then represents each simplified geometry compactly, so that the storage size is small while retrieving CDC relations from the representation is still reasonably fast. More specifically, the method called necessary cut is used to construct simple geometries , and the two methods, viz. the polygon representation and the rectangle representation , are devised to compactly represent the constructed geometries in cubic time w.r.t. the size of the corresponding simple geometry. Theoretical analyses demonstrate that the two representations, especially the rectangle representation, are promising to have small storage size. Moreover, our empirical evaluations on real-world datasets show that, for each dataset the new approach can produce a rectangle representation that has dominant performance against the state of the art techniques in reducing the storage size of the relations, while the average efficiency of retrieving CDC relations based on the rectangle representation is about the same as the fastest method in the literature.

Emergency Medical Services (EMS) are of great importance to saving people's lives from emergent accidents and diseases by efficiently picking up patients using ambulances. The transporting capability of an EMS system (e.g., defined as the average pickup time of patients) significantly depends on the real-time redeployment strategy of ambulances. That is, which station should an ambulance be redeployed to, after it becomes available (after it transports a patient to a hospital or after it finishes the in-site treatment for a patient)? However, it is a challenging task concerning with the multiple data D1-D5 as detailed in Introduction. To this end, in this paper, we propose a data-driven real-time ambulance redeployment approach that redeploys an ambulance to a proper station after it becomes available, so as to optimize the transporting capability of an EMS system, considering the aforementioned multiple data D1-D5. Specifically, the proposed approach is comprised of two stages to well consider the D1-D5. First, we propose a method (a safety time-based urgency index) to incorporate D1, D2, and D3 into each ambulance station's urgency degree (D*). Second, we propose an optimal matching algorithm to combine D*, D4, and D5 into the redeployment of the current available ambulance. Experimental results using data collected in real world demonstrate the significant advantages of our approach over many baselines. Comparing with baselines, our approach can save ~4 minutes (~35 percent) of the average pickup time for each patient, improve the ratio of patients picked up within 10 minutes from 0.684 and 0.803 (~17 percent), and largely enhance the survival rate of patients (~12 percent for patients in category A1 and ~17 percent for patients in A2).

This paper focuses on two related subtasks of aspect-based sentiment analysis, namely aspect term extraction and aspect sentiment classification, which we call aspect term-polarity co-extraction.The former task is to extract aspects of a product or service from an opinion document, and the latter is to identify the polarity expressed in the document about these extracted aspects.Most existing algorithms address them as two separate tasks and solve them one by one, or only perform one task, which can be complicated for real applications.In this paper, we treat these two tasks as two sequence labeling problems and propose a novel Dual crOss-sharEd RNN framework (DOER) to generate all aspect termpolarity pairs of the input sentence simultaneously.Specifically, DOER involves a dual recurrent neural network to extract the respective representation of each task, and a cross-shared unit to consider the relationship between them.Experimental results demonstrate that the proposed framework outperforms state-of-the-art baselines on three benchmark datasets.

With the increasing vehicles on the road, it is becoming more and more important to sense citywide traffic, which is of great benefit to the government's policy-making and people's decision making. Currently, traffic speed and volume information are mostly derived from GPS trajectories data and volume sensor records respectively. Unfortunately, speed and volume information suffer from serious data missing problem. Speed can be absent at arbitrary road segment and time slot, while volume is only recorded by limited volume sensors. For modeling citywide traffic, inspired by the observations of missing patterns and prior knowledge about traffic, we propose a neural memorization and generalization approach to infer the missing speed and volume, which mainly consists of a memorization module for speed inference and a generalization module for volume inference. Considering the temporal closeness and period properties, memorization module takes advantage of neural multi-head self-attention architecture to memorize the intrinsic correlations from historical traffic information. Generalization module adopts neural key-value attention architecture to generalize the extrinsic dependencies among volume sensors by exploiting road contexts. We conduct extensive experiments on two real-world datasets in two cities, Guiyang and Jinan, and the experimental results consistently demonstrate the advantages of our approach. We have developed a real-time system on the cloud, entitled CityTraffic, providing citywide traffic speed and volume information and fine-grained pollutant emission of vehicles in Guiyang city.

Hao Wang, Bing Liu, Chaozhuo Li, Yan Yang, Tianrui Li. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Traffic flow prediction has been regarded as a key research problem in the intelligent transportation system. In this paper, we propose an encoder-decoder model with temporal attention mechanism for multi-step forward traffic flow prediction task, which uses LSTM as the encoder and decoder to learn the long dependencies features and nonlinear characteristics of multivariate traffic flow related time series data, and also introduces a temporal attention mechanism for more accurately traffic flow prediction. Through the real traffic flow dataset experiments, it has shown that the proposed model has better prediction ability than classic shallow learning and baseline deep learning models. And the predicted traffic flow value can be well matched with the ground truth value not only under short step forward prediction condition but also under longer step forward prediction condition, which validates that the proposed model is a good option for dealing with the realtime and forward-looking problems of traffic flow prediction task.

In this paper, we present a road crack segmentation method based on generative adversarial networks (GAN). Our GAN networks consist of two neural network models in terms of a generator and a discriminator, where two improved networks CU-Net and FU-Net are proposed based on U-Net. The U-Net, CU-Net and FU-Net are used as the generator, while two-class networks are used as the discriminator. The purpose of using the generator is to generate fake crack images which are very similar to real crack images. And the recognition is done by the discriminator to distinguish the real crack images from the fake crack images. After iterative training between the generator and the discriminator, the generator can generate fake crack images that are very similar to the real crack image. Finally, the generator can be used to segment the road crack images. Compared with the other state-of-the-art methods on three datasets, the proposed method achieves better performance. Specifically, the precision, recall and F1-score are 91.46%, 73.40%, and 77.33%, respectively on one of the public datasets.

Accurate and effective prediction of future traffic accident risk is critical to reducing the number of traffic accidents, which is also of great help to personal safe travel. In our paper, we choose the real traffic administrative area as the way of regional division rather than grid map, so that our prediction results can be applied to true traffic scenarios directly. Instead of considering traffic flow as a single factor affecting traffic accidents, we divide traffic flow into multiple traffic volumes based on vehicle type. In order to better model the dynamic impact of different traffic flow data and traffic accident data in the local region and global regions for future traffic accident risk prediction, we design a deep learning framework to predict regional Traffic Accident risk that utilizes a Spatial-Temporal Attention Network (named TA-STAN). We also integrate many external environmental factors to further improve the accuracy. We evaluate our TA-STAN model on the real traffic accident dataset in New York City. The experimental results show that TA-STAN outperforms 6 baseline models in 3 evaluation metrics. More importantly, by visualizing the weight of attention, we can reasonably interpret the actual meaning of attention weights, which plays a crucial role in our model.

Weather forecasting is usually solved through numerical weather prediction (NWP), which can sometimes lead to unsatisfactory performance due to inappropriate setting of the initial states. In this paper, we design a data-driven method augmented by an effective information fusion mechanism to learn from historical data that incorporates prior knowledge from NWP. We cast the weather forecasting problem as an end-to-end deep learning problem and solve it by proposing a novel negative log-likelihood error (NLE) loss function. A notable advantage of our proposed method is that it simultaneously implements single-value forecasting and uncertainty quantification, which we refer to as deep uncertainty quantification (DUQ). Efficient deep ensemble strategies are also explored to further improve performance. This new approach was evaluated on a public dataset collected from weather stations in Beijing, China. Experimental results demonstrate that the proposed NLE loss significantly improves generalization compared to mean squared error (MSE) loss and mean absolute error (MAE) loss. Compared with NWP, this approach significantly improves accuracy by 47.76%, which is a state-of-the-art result on this benchmark dataset.

Ordering take-out food (a.k.a. takeaway food) on online-to-offline (O2O) food ordering and delivery platforms is becoming a new lifestyle for people living in big cities, thanks to its great convenience. Web users and mobile device users can order take-out food (i.e. obtain online food ordering services) on an O2O platform. Then the O2O platform will dispatch food carriers to deliver food from restaurants to users, i.e. providing users with offline food delivery services. For an O2O food ordering and delivery platform, improving food delivery efficiency, given the massive number of food orders each day and the limited number of food carriers, is of paramount importance to reducing the length of time users wait for their food. Thus, in this paper, we study the food delivery task grouping problem so as to improve food delivery efficiency and alleviate the pain of waiting for users, which to the best of our knowledge has not been studied yet. However, the food delivery task grouping problem is challenging, given two reasons. First, the food delivery efficiency is affected by multiple factors, which are non-trivial to formulate and jointly consider. Second, the problem is a typical NP-hard problem and to find near-optimal grouping results is not easy. To address these two issues, we propose an effective task grouping method. On one hand, we provide formal formulations for the factors affecting the food delivery efficiency, and provide an objective to organically combine these factors such that it can better guide the task grouping. On the other hand, we propose heuristic algorithms to efficiently obtain effective task grouping results, consisting of a greedy algorithm and a replacement algorithm. We evaluate our task grouping method using take-out food order data from web users and mobile device users on a real-world O2O food ordering and delivery platform. Experiment results demonstrate that our task grouping method can save ~ 16% (87 seconds) of average waiting time for each user, comparing with many baseline methods. It indicates that our method is able to significantly improve the food delivery efficiency and can provide better food delivery services for users.

Aspect term extraction is one of the important subtasks in aspect-based sentiment analysis. Previous studies have shown that using dependency tree structure representation is promising for this task. However, most dependency tree structures involve only one directional propagation on the dependency tree. In this paper, we first propose a novel bidirectional dependency tree network to extract dependency structure features from the given sentences. The key idea is to explicitly incorporate both representations gained separately from the bottom-up and top-down propagation on the given dependency syntactic tree. An end-to-end framework is then developed to integrate the embedded representations and BiLSTM plus CRF to learn both tree-structured and sequential features to solve the aspect term extraction problem. Experimental results demonstrate that the proposed model outperforms state-of-the-art baseline models on four benchmark SemEval datasets.

Collaborative clustering aims to find a common data structure between several distributed data sets governed by different privacy constraints and technical limitations that prohibit a central collection of data for processing. Therefore, it is required to process the data sets separately using collaboration, which allows clustering algorithms to work locally on an individual data set while exchanging information about the finding with algorithms in other data locations. Thus, the different data locations share information to improve individual clustering result amidst technical and privacy limitations but without breaching privacy. In this article, we present a framework of collaborative clustering that does not require interaction coefficients to regulate the effect of collaboration. We further adapt the framework to cluster distributed data using crisp and fuzzy clustering algorithms. We use particle swarm optimization techniques to inference the framework and, therefore, call it particle subswarms. Moreover, the collaboration increases the number of particles in the swarm without increasing the number of clusters in the data set. This article, therefore, provides the theoretical foundations of particle subswarms and some experimental results on several data sets.

Restricted Boltzmann machines (RBMs) and their variants are usually trained by contrastive divergence (CD) learning, but the training procedure is an unsupervised learning approach, without any guidances of the background knowledge. To enhance the expression ability of traditional RBMs, in this paper, we propose pairwise constraints (PCs) RBM with Gaussian visible units (pcGRBM) model, in which the learning procedure is guided by PCs and the process of encoding is conducted under these guidances. The PCs are encoded in hidden layer features of pcGRBM. Then, some pairwise hidden features of pcGRBM flock together and another part of them are separated by the guidances. In order to deal with real-valued data, the binary visible units are replaced by linear units with Gaussian noise in the pcGRBM model. In the learning process of pcGRBM, the PCs are iterated transitions between visible and hidden units during CD learning procedure. Then, the proposed model is inferred by approximative gradient descent method and the corresponding learning algorithm is designed. In order to compare the availability of pcGRBM and traditional RBMs with Gaussian visible units, the features of the pcGRBM and RBMs hidden layer are used as input "data" for K -means, spectral clustering (SP) and affinity propagation (AP) algorithms, respectively. We also use tenfold cross-validation strategy to train and test pcGRBM model to obtain more meaningful results with PCs which are derived from incremental sampling procedures. A thorough experimental evaluation is performed with 12 image datasets of Microsoft Research Asia Multimedia. The experimental results show that the clustering performance of K -means, SP, and AP algorithms based on pcGRBM model are significantly better than traditional RBMs. In addition, the pcGRBM model for clustering tasks shows better performance than some semi-supervised clustering algorithms.

Conventional haze-removal methods are designed to adjust the contrast and saturation, and in so doing enhance the quality of the reconstructed image. Unfortunately, the removal of haze in this manner can shift the luminance away from its ideal value. In other words, haze removal involves a tradeoff between luminance and contrast. We reformulated the problem of haze removal as a luminance reconstruction scheme, in which an energy term is used to achieve a favorable tradeoff between luminance and contrast. The proposed method bases the luminance values for the reconstructed image on statistical analysis of haze-free images, thereby achieving contrast values superior to those obtained using other methods for a given brightness level. We also developed a novel module for the estimation of atmospheric light using the color constancy method. This module was shown to outperform existing methods, particularly when noise is taken into account. The proposed framework requires only 0.55 seconds to process a 1-megapixel image. Experimental results demonstrate that the proposed haze-removal framework conforms to our theory of contrast.

This article aims to develop a new and robust approach to feature representation. Motivated by the success of Auto-Encoders, we first theoretically analyze and summarize the general properties of all algorithms that are based on traditional Auto-Encoders: (1) The reconstruction error of the input cannot be lower than a lower bound, which can be viewed as a guiding principle for reconstructing the input. Additionally, when the input is corrupted with noises, the reconstruction error of the corrupted input also cannot be lower than a lower bound. (2) The reconstruction of a hidden representation achieving its ideal situation is the necessary condition for the reconstruction of the input to reach the ideal state. (3) Minimizing the Frobenius norm of the Jacobian matrix of the hidden representation has a deficiency and may result in a much worse local optimum value. We believe that minimizing the reconstruction error of the hidden representation is more robust than minimizing the Frobenius norm of the Jacobian matrix of the hidden representation. Based on the above analysis, we propose a new model termedDouble Denoising Auto-Encoders(DDAEs), which uses corruption and reconstruction on both the input and the hidden representation. We demonstrate that the proposed model is highly flexible and extensible and has a potentially better capability to learn invariant and robust feature representations. We also show that our model is more robust than Denoising Auto-Encoders (DAEs) for dealing with noises or inessential features. Furthermore, we detail how to train DDAEs with two different pretraining methods by optimizing the objective function in a combined and separate manner, respectively. Comparative experiments illustrate that the proposed model is significantly better for representation learning than the state-of-the-art models.

The residual network has witnessed a great success in computer vision particularly on classification tasks, however, it has not been well studied in regression. In this work, we show its competence in a regression task - crowds flow prediction, which has strong implication to city safety and management. The problem of crowds flow prediction is challenging due to its fast dynamics. To address this issue, we explore residual learning with Gaussian regularization and propose a novel convolutional neural network called Gaussian noise residual networks (Noise-ResNet). Compared with the benchmark ST-ResNet on crowds flow prediction, the proposed architecture has three advantages: 1) Superior performance. Especially, it attains the state-of-the-art results on benchmark dataset BikeNYC. 2) Light architecture. Noise-ResNet only utilises one residual unit rather than STResNet with multiple ones, which greatly reduces the training time. 3) Interpretable input sequences. Noise-ResNet takes an input sequence that only considers the most important periodic data and closeness data, which makes the learning process more interpretable. Furthermore, experimental results substantiate that the Noise-ResNet can outperform ResNet with dropout on the same regression task.

Accompanying the rapid urbanization, many developing countries are suffering from serious air pollution problem. The demand for predicting future air quality is becoming increasingly more important to government's policy-making and people's decision making. In this paper, we predict the air quality of next 48 hours for each monitoring station, considering air quality data, meteorology data, and weather forecast data. Based on the domain knowledge about air pollution, we propose a deep neural network (DNN)-based approach (entitled DeepAir), which consists of a spatial transformation component and a deep distributed fusion network. Considering air pollutants' spatial correlations, the former component converts the spatial sparse air quality data into a consistent input to simulate the pollutant sources. The latter network adopts a neural distributed architecture to fuse heterogeneous urban data for simultaneously capturing the factors affecting air quality, e.g. meteorological conditions. We deployed DeepAir in our AirPollutionPrediction system, providing fine-grained air quality forecasts for 300+ Chinese cities every hour. The experimental results on the data from three-year nine Chinese-city demonstrate the advantages of DeepAir beyond 10 baseline methods. Comparing with the previous online approach in AirPollutionPrediction system, we have 2.4%, 12.2%, 63.2% relative accuracy improvements on short-term, long-term and sudden changes prediction, respectively.