Time series forecasting requires architectures that simultaneously achieve three competing objectives: (1) strict temporal causality for reliable predictions, (2) sub-quadratic complexity for practical scalability, and (3) multi-scale pattern recognition for accurate long-horizon forecasting. We introduce AutoHFormer, a hierarchical autoregressive transformer that addresses these challenges through three key innovations: 1) Hierarchical Temporal Modeling: Our architecture decomposes predictions into segment-level blocks processed in parallel, followed by intra-segment sequential refinement. This dual-scale approach maintains temporal coherence while enabling efficient computation. 2) Dynamic Windowed Attention: The attention mechanism employs learnable causal windows with exponential decay, reducing complexity while preserving precise temporal relationships. This design avoids both the anti-causal violations of standard transformers and the sequential bottlenecks of RNN hybrids. 3) Adaptive Temporal Encoding: a novel position encoding system is adopted to capture time patterns at multiple scales. It combines fixed oscillating patterns for short-term variations with learnable decay rates for long-term trends. Comprehensive experiments demonstrate that AutoHFormer 10.76X faster training and 6.06X memory reduction compared to PatchTST on PEMS08, while maintaining consistent accuracy across 96-720 step horizons in most of cases. These breakthroughs establish new benchmarks for efficient and precise time series modeling. Implementations of our method and all baselines in hierarchical autoregressive mechanism are available at https://github.com/lizzyhku/Autotime.

Sequential recommendation systems require both temporal efficiency to handle long interaction histories and hierarchical representation to model complex user–item relationships. Existing approaches face a fundamental tension: Mamba-based methods offer linear-time efficiency ( \(\mathcal{O}(L)\) ) but operate in Euclidean space, which distorts hierarchical patterns; hyperbolic models capture taxonomies well but suffer quadratic complexity ( \(\mathcal{O}(L^{2})\) ). To solve this dual challenge , we propose Hyperbolic Mamba (HMamba), the first architecture that unifies curvature-aware state spaces with hyperbolic geometry. Our key insight is that hyperbolic curvature \(\kappa\) simultaneously governs: (1) state transition granularity through \(\mathbf{\bar{A}}=\exp(\Delta\mathbf{A}\odot\mathbf{K}(\kappa))\) and (2) hierarchical distance preservation via \(d_{\mathcal{L}}\propto\sqrt{\kappa}\log(\cdot)\) . This enables joint optimization of efficiency and hierarchy—addressing the previously unsolved problem of deep-long modeling . Experiments show HMamba achieves 3–11% accuracy gains while maintaining 3.2 \(\times\) faster training than attention-based models, establishing a new paradigm for hierarchy-aware sequential recommendation. The code and datasets accompanying our paper are publicly available at https://github.com/CoderPowerBeyond/HMamba .

Spatial-temporal graph representations play a crucial role in urban sensing applications, including traffic analysis, human mobility behavior modeling, and citywide crime prediction. However, a key challenge lies in the noisy and sparse nature of spatial-temporal data, which limits existing neural networks' ability to learn meaningful region representations in the spatial-temporal graph. To overcome these limitations, we propose HGAurban, a novel heterogeneous spatial-temporal graph masked autoencoder that leverages generative self-supervised learning for robust urban data representation. Our framework introduces a spatial-temporal heterogeneous graph encoder that extracts region-wise dependencies from multi-source data, enabling comprehensive modeling of diverse spatial relationships. Within our self-supervised learning paradigm, we implement a masked autoencoder that jointly processes node features and graph structure. This approach automatically learns heterogeneous spatial-temporal patterns across regions, significantly improving the representation of dynamic temporal correlations. Comprehensive experiments across multiple spatiotemporal mining tasks demonstrate that our framework outperforms state-of-the-art methods and robustly handles real-world urban data challenges, including noise and sparsity in both spatial and temporal dimensions.

Achieving counterfactual fairness is a critical objective in advancing fairness research within machine learning. Studies have shown that machine learning models often inherit biases from their training data, leading to unfair decision-making. Fair data generation methods aim to mitigate these biases, ensuring that predictors trained on such data uphold fairness. However, in the context of counterfactual fairness, existing methods for generating fair data are often limited in their applicability and lead to significant performance losses in downstream predictors. To address these issues, this paper proposes a new algorithm for generating counterfactually fair data, allowing predictors trained on this generated data to adhere to counterfactual fairness. We propose a new metric, Intervention-Driven Correlation (IDC), to evaluate the counterfactual fairness of generative models. IDC assesses fairness by applying random interventions to samples and measuring the statistical correlation between the degree of intervention and the outcome of interest. This metric is applicable to both discrete and continuous sensitive attributes and labels. Furthermore, our studies reveal a critical insight: counterfactually fair data does not always guarantee counterfactually fair predictors when deployed in real-world scenarios. We identify the root causes of this issue and propose a robust solution. To bridge this gap, we propose the IDC-Reduction method, which ensures the fairness of downstream predictors by generating counterfactually fair data. Experimentally, our method outperforms existing approaches and achieves counterfactual fairness regardless of the type of downstream predictors.

The widespread adoption of smartphones and Location-Based Social Networks has led to a massive influx of spatio-temporal data, creating unparalleled opportunities for enhancing Point-of-Interest (POI) recommendation systems. These advanced POI systems are crucial for enriching user experiences, enabling personalized interactions, and optimizing decision-making processes in the digital landscape. However, existing surveys tend to focus on traditional approaches and few of them delve into cutting-edge developments, emerging architectures, as well as security considerations in POI recommendations. To address this gap, our survey stands out by offering a comprehensive, up-to-date review of POI recommendation systems, covering advancements in models, architectures, and security aspects. We systematically examine the transition from traditional models to advanced techniques such as large language models. Additionally, we explore the architectural evolution from centralized to decentralized and federated learning systems, highlighting the improvements in scalability and privacy. Furthermore, we address the increasing importance of security, examining potential vulnerabilities and privacy-preserving approaches. Our taxonomy provides a structured overview of the current state of POI recommendation, while we also identify promising directions for future research in this rapidly advancing field.

With the widespread development of database systems, data security has become crucial when it comes to sharing among users and servers. A straightforward approach involves using searchable encryption to ensure the confidentiality of shared data. However, in certain scenarios, varying user tiers are granted disparate data searching privileges, and administrators need to restrict the searchability of ciphertexts to select users exclusively. To address this issue, public key encryption with authorized keyword search (PEAKS) was proposed, wherein solely authorized users possess the ability to conduct targeted keyword searches. Nonetheless, it is vulnerable to resist quantum computing attacks. As a result, research focusing on authorizing users to search for keywords while achieving quantum security is far-reaching. In this paper, we propose a lattice-based variant of PEAKS (L-PEAKS) that enables keyword dataset authorization for outsourced data management. Unlike existing schemes, our design incorporates identity-based encryption (IBE) to overcome the bottleneck of public key management. Besides, we utilize several lattice sampling algorithms to defend against attacks from quantum adversaries. Specifically, each authorized user must obtain a search privilege from an authority. The authority distributes an authorized token to the user within a specific time period, and the user generates a trapdoor for any authorized keywords. Our scheme is proven to be secure against IND-sID-CKA and T-EUF security in a quantum setting. We also conduct comprehensive evaluations on a commodity machine to assess completeness and provide theoretical complexity comparisons with existing state-of-the-art schemes.

Graph augmentation with contrastive learning has gained significant attention in the field of recommendation systems due to its ability to learn expressive user representations, even when labeled data is limited. However, directly applying existing GCL models to real-world recommendation environments poses challenges. There are two primary issues to address. Firstly, the lack of consideration for data noise in contrastive learning can result in noisy self-supervised signals, leading to degraded performance. Secondly, many existing GCL approaches rely on graph neural network (GNN) architectures, which can suffer from over-smoothing problems due to non-adaptive message passing. To address these challenges, we propose a principled framework called GraphAug. This framework introduces a robust data augmentor that generates denoised self-supervised signals, enhancing recommender systems. The GraphAug framework incorporates a graph information bottleneck (GIB)-regularized augmentation paradigm, which automatically distills informative self-supervision information and adaptively adjusts contrastive view generation. Through rigorous experimentation on real-world datasets, we thoroughly assessed the performance of our novel GraphAug model. The outcomes consistently unveil its superiority over existing baseline methods. The source code for our model is publicly available at: https://github.com/HKUDS/GraphAug.

Behemoth graphs are often fragmented and separately stored by multiple data owners as distributed subgraphs in many realistic applications. Without harming data privacy, it is natural to consider the subgraph federated learning (subgraph FL) scenario, where each local client holds a subgraph of the entire global graph, to obtain globally generalized graph mining models. To overcome the unique challenge of incomplete information propagation on local subgraphs due to missing cross-subgraph neighbors, previous works resort to the augmentation of local neighborhoods through the joint FL of missing neighbor generators and GNNs. Yet their technical designs have profound limitations regarding the utility, efficiency, and privacy goals of FL. In this work, we propose FedDEP to comprehensively tackle these challenges in subgraph FL. FedDEP consists of a series of novel technical designs: (1) Deep neighbor generation through leveraging the GNN embeddings of potential missing neighbors; (2) Efficient pseudo-FL for neighbor generation through embedding prototyping; and (3) Privacy protection through noiseless edge-local-differential-privacy. We analyze the correctness and efficiency of FedDEP, and provide theoretical guarantees on its privacy. Empirical results on four real-world datasets justify the clear benefits of proposed techniques.

Nowadays, it becomes a common practice to capture some data of sports games with devices such as GPS sensors and cameras and then use the data to perform various analyses on sports games, including tactics discovery, similar game retrieval, performance study, and so forth. While this practice has been conducted to many sports such as basketball and soccer, it remains largely unexplored on the billiards sports, which is mainly due to the lack of publicly available datasets. Motivated by this, we collect a dataset of billiards sports, which includes the layouts (i.e., locations) of billiards balls after performing break shots, called break shot layouts, the traces of the balls as a result of strikes (in the form of trajectories), and detailed statistics and performance indicators. We then study and develop techniques for three tasks on the collected dataset, including (1) prediction and (2) generation on the layouts data, and (3) similar billiards layout retrieval on the layouts data, which can serve different users such as coaches, players and fans. We conduct extensive experiments on the collected dataset and the results show that our methods perform effectively and efficiently.

Using NLP to analyze ESG reports has gained a lot of attention. However, existing supervised learning approaches rely on high-level and predetermined ESG topics (as used by reporting standards/rating agencies), which often fail to capture specific, latest trends and impactful issues in specific industries, while fully unsupervised approaches yield generic topics that are not useful for practical analysis. We proposed a novel data-driven and dynamic approach that base on the report contents to identify important and trendy issues that cannot be revealed by previous approaches. Technically speaking, our approach combines supervised text classification on industry-specific material topics with unsupervised topic modeling. The identified issues can be ranked using a simple word counting method. To illustrate the usefulness of our methodology, we apply it to a set of ESG reports from the banking industry. The identified issues, representing the trendy issues, can also be used to show the different priorities of focuses between banks from different regions. Time-series analysis can be done as well to see the changes in priority of issues over time. We are able to validate (indirectly and intuitively) that some of the issues should be correct, which show that our approach is promising.

Detecting anomalous trajectories has become an important task in many location-based applications. While many approaches have been proposed for this task, they suffer from various issues including (1) incapability of detecting anomalous subtrajectories, which are finer-grained anomalies in trajectory data, and/or (2) non-data driven, and/or (3) requirement of sufficient supervision labels which are costly to collect. In this paper, we propose a novel reinforcement learning based solution called RL4OASD, which avoids all aforementioned issues of existing approaches. RL4OASD involves two networks, one responsible for learning features of road networks and trajectories and the other responsible for detecting anomalous subtrajectories based on the learned features, and the two networks can be trained iteratively without labeled data. Extensive experiments are conducted on two real datasets, and the results show that our solution can significantly outperform the state-of-the-art methods (with 20-30% improvement) and is efficient for online detection (it takes less than 0.1ms to process each newly generated data point).

Among various region embedding methods, graph-based region relation learning models stand out, owing to their strong structure representation ability for encoding spatial correlations with graph neural networks. Despite their effectiveness, several key challenges have not been well addressed in existing methods: i) Data noise and missing are ubiquitous in many spatio-temporal scenarios due to a variety of factors. ii) Input spatio-temporal data (e.g., mobility traces) usually exhibits distribution heterogeneity across space and time. In such cases, current methods are vulnerable to the quality of the generated region graphs, which may lead to suboptimal performance. In this paper, we tackle the above challenges by exploring the Automated Spatio-Temporal graph contrastive learning paradigm (AutoST) over the heterogeneous region graph generated from multi-view data sources. Our AutoST framework is built upon a heterogeneous graph neural architecture to capture the multi-view region dependencies with respect to POI semantics, mobility flow patterns and geographical positions. To improve the robustness of our GNN encoder against data noise and distribution issues, we design an automated spatio-temporal augmentation scheme with a parameterized contrastive view generator. AutoST can adapt to the spatio-temporal heterogeneous graph with multi-view semantics well preserved. Extensive experiments for three downstream spatio-temporal mining tasks on several real-world datasets demonstrate the significant performance gain achieved by our AutoST over a variety of baselines. The code is publicly available at https://github.com/HKUDS/AutoST.

With the proliferation of tracking devices such as cameras and/or GPS sensors, sports data is being generated at an unprecedented speed and the interest in collecting some data from sports games has grown dramatically as well. The collected data facilitates various sports analytic tasks; however, these studies are mainly concerning with sports such as football and basketball. It remains largely unexplored for billiards sports though it is a popular sport of both strategy and physical skill, and this is mainly due to the lack of publicly available datasets. Motivated by this, we collect a dataset of billiards sports, which includes the layouts (i.e., locations) of billiards balls after performing break shots, called break shot layouts, the traces of the balls as a result of strikes (in the form of trajectories), and detailed statistics and performance indicators. On top of the dataset, we investigate several tasks, including prediction and generation on the layouts data and similarity search on the trajectory data, which can serve different users such as coaches, players and fans. We conduct extensive experiments on the collected dataset for the tasks, and the results demonstrate the superior performance of the methods proposed in this paper.

We address security issues in a cloud database system which employs the DBaaS model --- a data owner (DO) exports data to a cloud database service provider (SP). To provide data security, sensitive data is encrypted by the DO before it is uploaded to the SP. Compared to existing secure query processing systems like CryptDB [7] and MONOMI [8], in which data operations (e.g., comparison or addition) are supported by specialized encryption schemes, our demo system, SDB, is implemented based on a set of data-interoperable secure operators, i.e., the output of an operator can be used as input of another operator. As a result, SDB can support a wide range of complex queries (e.g., all TPC-H queries) efficiently. In this demonstration, we show how our SDB prototype supports secure query processing on complex workload like TPC-H. We also demonstrate how our system protects sensitive information from malicious attackers.

Bayesian networks (BNs) are popular for modeling conditional distributions of variables and causal relationships, especially in biological settings such as protein interactions, gene regulatory networks and microbial interactions. Previous BN structure learning algorithms treat variables with similar tendency separately. In this paper, we propose a grouped sparse Gaussian BN (GSGBN) structure learning algorithm which creates BN based on three assumptions: (i) variables follow a multivariate Gaussian distribution, (ii) the network only contains a few edges (sparse), (iii) similar variables have less-divergent sets of parents, while not-so-similar ones should have divergent sets of parents (variable grouping). We use L1regularization to make the learned network sparse, and another term to incorporate shared information among variables. For similar variables, GSGBN tends to penalize the differences of similar variables' parent sets more, compared to those not-so-similar variables' parent sets. The similarity of variables is learned from the data by alternating optimization, without prior domain knowledge. Based on this new definition of the optimal BN, a coordinate descent algorithm and a projected gradient descent algorithm are developed to obtain edges of the network and also similarity of variables. Experimental results on both simulated and real datasets show that GSGBN has substantially superior prediction performance for structure learning when compared to several existing algorithms.

We address security issues in a cloud database system which employs the DBaaS model. In such a model, a data owner (DO) exports its data to a cloud database service provider (SP). To provide data security, sensitive data is encrypted by the DO before it is uploaded to the SP. Existing encryption schemes, however, are only partially homomorphic in the sense that each of them was designed to allow one specific type of computation to be done on encrypted data. These existing schemes cannot be integrated to answer real practical queries that involve operations of different kinds. We propose and analyze a secure query processing system (SDB) on relational tables and a set of elementary operators on encrypted data that allow data interoperability, which allows a wide range of SQL queries to be processed by the SP on encrypted information. We prove that our encryption scheme is secure against two types of threats and that it is practically efficient.

Peer Data Management Systems (PDMS) are an attractive solution for managing distributed heterogeneous information. When a peer (client) requests data from another peer (server) with a different schema, translations of the query and its answer are done by a sequence of intermediate peers (translators). There are two privacy issues in this P2P data integration process: (i) answer privacy: no unauthorized parties (including the translators) should learn the query result; (ii) mapping privacy: the schema and the value mappings used by the translators to perform the translation should not be revealed to other peers. Elmeleegy and Ouzzani proposed the PPP protocol that is the first to support privacy-preserving querying in PDMS. However, PPP suffers from several shortcomings. First, PPP does not satisfy the requirement of answer privacy, because it is based on commutative encryption; we show that this issue can be fixed by adopting another cryptographic technique called oblivious transfer. Second, PPP adopts a weaker notion for mapping privacy, which allows the client peer to observe certain mappings done by translators. In this paper, we develop a lightweight protocol, which satisfies mapping privacy and extend it to a more complex one that facilitates parallel translation by peers. Furthermore, we consider a stronger adversary model where there may be collusions among peers and propose an efficient protocol that guards against collusions. We conduct an experimental study on the performance of the proposed protocols using both real and synthetic data. The results show that the proposed protocols not only achieve a better privacy guarantee than PPP, but they are also more efficient.

Queries on semistructured data are hard to process due to the complex nature of the data and call for specialized techniques. Existing path-based indexes and query processing algorithms are not efficient for searching complex structures beyond simple paths, even when the queries are high-selective. We introduce the definition of minimal infrequent structures (MIS), which are structures that 1) exist in the data, 2) are not frequent with respect to a support threshold, and 3) all substructures of them are frequent. By indexing the occurrences of MIS, we can efficiently locate the high-selective substructures of a query, improving search performance significantly. An efficient data mining algorithm is proposed, which finds the minimal infrequent structures. Their occurrences in the XML data are then indexed by a lightweight data structure and used as a fast filter step in query evaluation. We validate the efficiency and applicability of our methods through experimentation on both synthetic and real data.

This paper investigates how to index a text which is subject to updates. The best solution in the literature (P.Ferragina, et al., 1998) is based on suffix tree using O(n log n) bits of storage, where n is the length of the text. It supports finding all occurrences of a pattern P in O(|P|+occ) time, where occ is the number of occurrences. Each text update consists of inserting or deleting a substring of length y and can be supported in O(y+/spl radic/(n)) time. In this paper, we initiate the study of compressed index using only O(n log |/spl Sigma/|) bits of space, where /spl Sigma/ denotes the alphabet. Our solution supports finding all occurrences of a pattern P in O(|P| Iog/sup 2/n(log/sup /spl epsi//n+log|/spl Sigma/|)+occlog/sup 1+/spl epsi//n) time, while insertion or deletion of a substring of length y can be done in O((y+/spl radic/(n)) Iog/sup 2+/spl epsi// n) amortized tune, where 0

With the standardization of XML as an information exchange language over the Internet, a huge amount of information is formatted in XML documents. In order to analyze this information efficiently, decomposing the XML documents and storing them in relational tables is a popular practice. However, query processing becomes expensive since, in many cases, an excessive number of joins is required to recover information from the fragmented data. If a collection consists of documents with different structures (for example, they come from different DTDs), mining clusters in the documents could alleviate the fragmentation problem. We propose a hierarchical algorithm (S-GRACE) for clustering XML documents based on structural information in the data. The notion of structure graph (s-graph) is proposed, supporting a computationally efficient distance metric defined between documents and sets of documents. This simple metric yields our new clustering algorithm which is efficient and effective, compared to other approaches based on tree-edit distance. Experiments on real data show that our algorithm can discover clusters not easily identified by manual inspection.

Approximate string matching on large DNA sequences data is very important in bioinformatics. Some studies have shown that suffix tree is an efficient data structure for approximate string matching. It performs better than suffix array if the data structure can be stored entirely in the memory. However our study find that suffix array is much better than suffix tree for indexing the DNA sequences since the data structure has to be created and stored on the disk due to its size. We propose a novel auxiliary data structure which greatly improves the efficiency of suffix array in the approximate string matching problem in the external memory model. The second problem we have tackled is the parallel approximate matching in DNA sequence. We propose 2 novel parallel algorithms for this problem and implement them on a PC cluster The result shows that when the error allowed is small, a direct partitioning of the array over the machines in the cluster is a more efficient approach. On the other hand, when the error allowed is large, partitioning the data over the machines is a better approach.