Cloud service providers face the challenge of determining optimal server replenishment policies that minimize inventory costs while ensuring the expected demand satisfaction rate. This article addresses a long-term, single-echelon inventory optimization problem tailored to the unique characteristics of cloud, incorporating varying given demand satisfaction rates for different cloud products. Traditional replenishment methods, based on statistical approaches or conventional reinforcement learning (RL) algorithms, are ill-suited due to the uncertainties in user demand retention and the diversity in demand satisfaction requirements. To overcome these limitations, we propose an enhanced replenishment policy under given demand satisfaction rates using RL, combined with our policy imitation. Specifically, we introduce a data-driven safety stock (DSS) model as an expert policy, utilizing Kullback-Leibler (KL) divergence and advantage values for effective action imitation. Based on the results of DSS, we design an efficient reward function and propose an approach for hyperparameter estimation. Together, they enable our method to learn better policies that align with the given demand satisfaction rate. We further design an end-to-end network that integrates demand retention prediction with RL, incorporating a learnable Gaussian kernel to model the lead time effects and local features, and a mean self-attention module for capturing global features. To accelerate training, we also introduce a parallel algorithm. Extensive experiments on real-world data from multiple internet data centers (IDCs), across different time scales and given demand satisfaction rates, demonstrate the superior performance of our method. Additionally, ablation studies and parameter sensitivity analyses verify the efficacy of our policy imitation method and network architecture.

AI coding agents are changing the bottleneck in software collaboration: code is increasingly cheap, while understanding intent, negotiating scope, and governing long-term project responsibility remain costly. This paper proposes Knowledge-Based Pull Requests (KPR), a trusted workflow for agent-mediated software collaboration across trust boundaries, including open source, enterprise, vendor, contractor, and customer-driven settings. In KPR, an external collaborator's local code, tests, and cleaned agent interaction trace are treated as knowledge sources rather than as the default merge candidate. Agents distill these sources into a human-confirmed knowledge package and render it into reviewer-facing forms such as design memos, risk checklists, test plans, or implementation briefs. A project-owned inner trusted coding agent then regenerates candidate code inside the receiving project's environment under repository context, engineering conventions, tests, and security policy. KPR therefore separates two decisions that traditional pull requests often collapse: whether the knowledge should enter the project, and whether a particular implementation should be merged. We contribute the KPR workflow, a candidate artifact schema, a cost-accounting view, a collaboration gateway architecture, a minimal controlled simulation pilot over seven merged public pull requests, and an evaluation agenda. The pilot shows that KPR packages can be instantiated from real PR material and stress-tested under description ablation, diff ablation, and synthetic poisoned-patch conditions. We position KPR as an empirically testable workflow: its value depends on whether auditable extraction, transformation, and project-side regeneration reduce the cost of understanding and reworking high-context external changes.

Reinforcement learning (RL) has shown promise for combinatorial optimization problems on graphs by learning heuristics that generalize across instances. However, effectively incorporating domain knowledge into RL frameworks for graph partitioning remains challenging, as existing approaches typically rely on unconstrained node-level actions that lead to large action spaces and inefficient exploration. In this paper, we propose RidgeCut, an RL framework that constrains the action space to enforce structure-aware partitioning in the Normalized Cut problem. Using transportation networks as a motivating example, we introduce a novel concept that leverages domain knowledge about urban road topology – where natural partitions often take the form of concentric rings and radial wedges. By transforming the graph into linear or circular representations, our method enables the use of transformer-based policies and efficient learning via Proximal Policy Optimization. The resulting partitions from RidgeCut are not only aligned with expected spatial layouts but also achieve lower normalized cuts compared to existing methods. Experimental results on synthetic and real-world traffic graphs demonstrate that RidgeCut consistently outperforms existing methods while exhibiting strong inductive generalization across graph sizes. Although motivated by road networks, RidgeCut provides a general mechanism for embedding structural priors into RL frameworks for graph partitioning.

The task of next-item recommendation is a crucial component in recommendation systems. The challenge of this task lies in extracting complex interaction information from users’ historical interactions with items. While prior research has transformed users’ interaction histories into graphs and hypergraphs to mine high-order interactions, the integration of both remains uncharted. Existing methods treat graphs and hypergraphs separately in representation learning, missing out on their shared attributes. Addressing this gap, we introduce a novel unified framework that integrates graphs and hypergraphs into one unified message passing paradigm. Our method employs graph neural networks and hypergraph neural networks to continuously conduct representation learning, achieving end-to-end next item recommendation. Specifically, our framework consists of the following components: 1) a unique method for graph and hypergraph construction from interaction histories, with items as nodes and users as hyperedges; 2) a novel message passing framework compatible with both graph and hypergraph neural networks; and 3) an innovative hypergraph neural network aggregation module enhanced with time and position encoders. The test results on multiple public benchmarks verify that our method outperforms current best practices in performance.

In real-world applications, GPS trajectories often suffer from low sampling rates, with large and irregular intervals between consecutive GPS points. This sparse characteristic presents challenges for their direct use in GPS-based systems. This paper addresses the task of map-constrained trajectory recovery, aiming to enhance trajectory sampling rates of GPS trajectories. Previous studies commonly adopt a sequence-to-sequence framework, where an encoder captures the trajectory patterns and a decoder reconstructs the target trajectory. Within this framework, effectively representing the road network and extracting relevant trajectory features are crucial for overall performance. Despite advancements in these models, they fail to fully leverage the complex spatio-temporal dynamics present in both the trajectory and the road network. To overcome these limitations, we categorize the spatio-temporal dynamics of trajectory data into two distinct aspects: spatial-temporal traffic dynamics and trajectory dynamics. Furthermore, We propose TedTrajRec, a novel method for trajectory recovery. To capture spatio-temporal traffic dynamics, we introduce PD-GNN, which models periodic patterns and learns topologically aware dynamics concurrently for each road segment. For spatio-temporal trajectory dynamics, we present TedFormer, a time-aware Transformer that incorporates temporal dynamics for each GPS location by integrating closed-form neural ordinary differential equations into the attention mechanism. This allows TedFormer to effectively handle irregularly sampled data. Extensive experiments on three real-world datasets demonstrate the superior performance of TedTrajRec. The code is publicly available at https://github.com/ysygMhdxw/TEDTrajRec/.

Hypergraph neural networks can model more flexible connectivity relationships, are used to model higher-order interactions, and have produced strong results in many real-world applications. However, the currently existing hypergraph neural networks need more exploration in capturing the global positional information of nodes in hypergraphs. Although there have been many explorations of the problem in graph neural networks, extending these approaches to hypergraphs is fraught with challenges. The major challenge is that hyperedges in hypergraphs are the other dimensional element of the incidence structure, have more flexible definitions than edges in graphs, and require more attention when learning global positional information. We propose a novel position-aware hypergraph message-passing neural network framework to address the above challenges. Specifically, we propose a global positional embedding learning approach that can separately model global positional information for nodes and hyperedges. At the same time, we also optimize the learning of local structures with hyperedges. Experiments on several publicly available benchmark datasets find that our proposed method outperforms many state-of-the-art methods.

Long-term time series forecasting (LTSF) is a fundamental task with wide-ranging applications. Although Transformer-based models have made significant breakthroughs in forecasting, their effectiveness for time series forecasting remains debatable. In this paper, we revisit the significance of self-attention and propose a simple yet effective mechanism, Periodic-Nested Group Attention, namely PENGUIN. Our approach highlights the importance of explicitly modeling periodic patterns and incorporating relative attention bias for effective time series modeling. To this end, we introduce a periodic-nested relative attention bias that captures periodic structures directly. To handle multiple coexisting periodicities (e.g., daily and weekly cycles), we design a grouped attention mechanism, where each group targets a specific periodicity using a multi-query attention mechanism. Extensive experiments across diverse benchmarks demonstrate that PENGUIN consistently outperforms both MLP-based and Transformer-based models.

Location-aware devices continuously generate massive volumes of trajectory data, creating demand for efficient compression. Line simplification is a common solution but typically assumes 2D trajectories and ignores time synchronization and motion continuity. We propose PILOT-C, a novel trajectory compression framework that integrates frequency-domain physics modeling with error-bounded optimization. Unlike existing line simplification methods, PILOT-C supports trajectories in arbitrary dimensions, including 3D, by compressing each spatial axis independently. Evaluated on four real-world datasets, PILOT-C achieves superior performance across multiple dimensions. In terms of compression ratio, PILOT-C outperforms CISED-W, the current state-of-the-art SED-based line simplification algorithm, by an average of 19.2%. For trajectory fidelity, PILOT-C achieves an average of 32.6% reduction in error compared to CISED-W. Additionally, PILOT-C seamlessly extends to three-dimensional trajectories while maintaining the same computational complexity, achieving a 49% improvement in compression ratios over SQUISH-E, the most efficient line simplification algorithm on 3D datasets.

Modeling route representation aims to obtain contextual representations of an entire route for various traffic-related tasks. In reality, spatial-temporal data often exhibits multi-scale characteristics, which are utilized by many studies to enhance their performance. However, there is still a lack of in-depth research on how to effectively incorporate the multi-scale spatial-temporal information into transformer structure to adequately model route representation. In this paper, we propose a novel hierarchical route representation framework called RouteMT, which effectively captures multi-scale spatial-temporal characteristics of routes and leverages a mixed-scale transformer architecture to fuse intra and interroute features. Experiments on real data confirm RouteMT’s superior performance and versatility.

Traffic speed prediction is vital for intelligent transportation systems. However, most existing methods focus on costly static sensors. In contrast, utilizing GPS devices from vehicles as mobile sensors offers a cost-effective means to gather dynamic traffic data. Despite the presence of historical trajectory data, mobile sensor-based traffic prediction remains under-explored. Existing methods often treat trajectories as substitutes for static sensors, missing the full utilization of the spatial-temporal signals within the complete trajectory set. To address this, we propose TrajHGT, a novel trajectory set empowered hypergraph transformer model that captures trafficrelated spatial-temporal features through adaptive attention and fusion mechanisms in both the trajectory hypergraph space and the road graph space. Real dataset experiments demonstrate the superiority of TrajHGT.

Accurately estimating route travel time is crucial for intelligent transportation systems. Urban road networks and routes can be viewed from spatial and topological perspectives while existing works typically focus on one view and disregard important information from the other perspective. In this paper, we propose, a novel travel time estimation model. It incorporates an alignment-enhanced spatial-topological aware dual transformer model to adaptively incorporate intra-and inter-view features in the route, guided by cross-view location alignment matrices with clear correspondences between locations in two views. Additionally, we propose a sparsity-aware dual-view traffic feature extraction module to effectively capture temporal traffic state changes. Compared to baseline models, demonstrates improved performance on the MAPE and MAE metrics for Chengdu and Shanghai datasets, achieving improvements of 8.32%, 7.03%, 8.06% and 9.51% respectively, validating the effectiveness of in travel time estimation.

Deep neural networks based on reinforcement learning (RL) for solving combinatorial optimization (CO) problems are developing rapidly and have shown a tendency to approach or even outperform traditional solvers. However, existing methods overlook an important distinction: CO problems differ from other traditional problems in that they focus solely on the optimal solution provided by the model within a specific length of time, rather than considering the overall quality of all solutions generated by the model. In this paper, we propose Leader Reward and apply it during two different training phases of the Policy Optimization with Multiple Optima (POMO) model to enhance the model's ability to generate optimal solutions. This approach is applicable to a variety of CO problems, such as the Traveling Salesman Problem (TSP), the Capacitated Vehicle Routing Problem (CVRP), and the Flexible Flow Shop Problem (FFSP), but also works well with other POMO-based models or inference phase's strategies. We demonstrate that Leader Reward greatly improves the quality of the optimal solutions generated by the model. Specifically, we reduce the POMO's gap to the optimum by more than 100 times on TSP100 with almost no additional computational overhead.

Adaptive traffic signal control plays a crucial role in enhancing the traffic situation in urban cities. Recently, reinforcement learning-based methods have demonstrated remarkable performance in addressing traffic signal control problems. However, the deployment of these methods in real-world environments is limited due to their lack of robustness. When errors occur during traffic data collection, their performance tends to deteriorate significantly. This paper addresses a prevalent issue where certain intersections, referred to as blinded intersections, cannot receive traffic data either due to hardware failures or the absence of traffic detectors. Existing adaptive traffic signal control methods fail to achieve satisfactory results in such scenarios as they are unable to learn effective policies for blinded intersections. Through theoretical analysis, we identify the primary reason behind the poor performance of existing methods as improper reward selection. To mitigate this issue, we propose a novel reward function called, which incorporates Number of Vehicles and Outflow, offering a correct optimization objective with low variance. These advantages significantly enhance the performance of reinforcement learning methods. Additionally, we introduce a new reinforcement learning model called BlindLight, which employs a dual model structure to learn Q-values for different types of intersections independently. This model design enhances the robustness of the system. Experimental results conducted on public datasets demonstrate the consistent performance improvement achieved by the theoretically supported reward function in existing methods with blinded intersections. Furthermore, BlindLight outperforms all state-of-the-art traffic signal control methods significantly.

GPS trajectories are the essential foundations for many trajectory-based applications. Most applications require a large number of high sample rate trajectories to achieve a good performance. However, many real-life trajectories are collected with low sample rate due to energy concern or other constraints. We study the task of trajectory recovery in this paper as a means to increase the sample rate of low sample trajectories. Most existing works on trajectory recovery follow a sequence-to-sequence diagram, with an encoder to encode a trajectory and a decoder to recover real GPS points in the trajectory. However, these works ignore the topology of road network and only use grid information or raw GPS points as input. Therefore, the encoder model is not able to capture rich spatial information of the GPS points along the trajectory, making the prediction less accurate and less spatial consistent. In this paper, we propose a road network enhanced transformer-based framework, namely RNTrajRec, for trajectory recovery. RNTrajRec first uses a graph model, namely GridGNN, to learn the embedding features of each road segment. It next develops a spatial-temporal transformer model, namely GPSFormer, to learn rich spatial and temporal features along with a Sub-Graph Generation module to capture the spatial features for each GPS point in the trajectory. It finally forwards the outputs of encoder model to a multi-task decoder model to recover the missing GPS points. Extensive experiments based on three large-scale real-life trajectory datasets confirm the effectiveness of our approach.

Hashing plays an important role in the content-based retrieval of multimedia data. Existing methods focus on designing various joint-optimization strategies to preserve the similarity relationships between different modalities and improve the performance of cross-modal retrieval. However, when the intrinsic information of any modality is significantly insufficient compared with others, the final hash space will collapse to the trivial space due to the joint training. Compared with inter-modal semantic alignment, directly aligning the semantics of each modality with the label-wise semantics can obtain higher-quality common semantic spaces, and only a pair-wise alignment between each space can obtain a unified representation. Following this idea, we design a general hash generation framework for uni-modal embedding and directly align the hashing code from different modalities using a pair-wise loss, further improving the retrieval performance within the less-semantic modality. Within this framework, we introduce two optimizations to penalize dissimilar items. First, with a Gaussian distribution to describe the hashing semantic distribution, JS-divergence is introduced to keep the consistency of label-wise semantics and hashing similarity. Then the attention mechanism is used for hard-sample re-weighting to learn fine-grained distribution alignment further. We conduct extensive experiments on three public datasets to validate the enhancements of our work.

Hamming space retrieval, allowing for retrieval within a fixed Hamming radius rather than scanning all instances linearly, has gained widespread attention for retrieving nearest neighbors at a fixed computational cost. Current models have focused only on shared-label correlations as the label-wise semantics while ignoring the potential multi-hop constraints, which are the unique constraints in Hamming space retrieval under multi-label conditions. Instances with multi-dimensional labels form a multi-hop correlation graph rather than several clusters in a single-label scenario. So there are potential distance constraints between dissimilar instances connected through multi-hop correlations. Existing models blindly expanding the dissimilar instances will break the multi-hop correlations and finally disrupt the hashing consistency of the shared-label instances. This paper first defines the multi-hop correlations to address these challenges with the multi-hop preserving strategy for dissimilar instances to help the Hamming distance converge to the expected range. Meanwhile, we introduce a Wasserstein-1-distance-based loss to reduce information loss while encoding compactly. We evaluate our proposed model on three commonly used datasets with extensive experiments to demonstrate that our model achieves significant improvements over existing methods.

Modeling continuous-time dynamics on irregular time series is critical to account for data evolution and correlations that occur continuously. Traditional methods including recurrent neural networks or Transformer models leverage inductive bias via powerful neural architectures to capture complex patterns. However, due to their discrete characteristic, they have limitations in generalizing to continuous-time data paradigms. Though neural ordinary differential equations (Neural ODEs) and their variants have shown promising results in dealing with irregular time series, they often fail to capture the intricate correlations within these sequences. It is challenging yet demanding to concurrently model the relationship between input data points and capture the dynamic changes of the continuous-time system. To tackle this problem, we propose ContiFormer that extends the relation modeling of vanilla Transformer to the continuous-time domain, which explicitly incorporates the modeling abilities of continuous dynamics of Neural ODEs with the attention mechanism of Transformers. We mathematically characterize the expressive power of ContiFormer and illustrate that, by curated designs of function hypothesis, many Transformer variants specialized in irregular time series modeling can be covered as a special case of ContiFormer. A wide range of experiments on both synthetic and real-world datasets have illustrated the superior modeling capacities and prediction performance of ContiFormer on irregular time series data. The project link is https://seqml.github.io/contiformer/.

Searching spatial objects is a fundamental task in spatial services such as online maps. Traditional search methods are based on filtering conditions, burdening users to specify their requirements. This paper focuses on spatial search via examples. Particularly, the user can specify an example, which is a set of objects of interest, and the purpose is to find a list of results, each containing a set of objects with similar properties to the given example. We conducted a user study, showing that a search interface based on examples can effectively complement existing approaches. However, the existing example-based search is not scalable, hindering its applications to larger datasets. To address this challenge, we propose two new algorithms, namely HSP and LORA, to efficiently answer example-based spatial queries. HSP is an algorithm based on a hierarchical partitioning of the search space, and it achieves up to 20 times faster than the state-of-the-art algorithm. LORA further improves the efficiency, running up to 5000 times faster than the state-of-the-art algorithm. We present a systematic evaluation to demonstrate the efficacy of our algorithms.

Consumer adoption of electric vehicles is essentially related to product quality factors, such as safety, performance and compatibility; however, the relationship between product quality standards and consumer behavior is not clear. Based on Multi-Attribute Utility Theory (MAUT) and Prospect Theory, we distinguish claimed quality attributes, intrinsic quality attributes, measured quality attributes and perceived quality attributes and establish a conceptional model using System Dynamics (SD) simulation from the perspective of a Standard-Information-Behavior framework to explore the heterogeneous impacts of technical standards on consumers’ willingness to adopt electric vehicles. Based on the theory model and simulation, we try to explain the heterogeneous effects of three different standards: safety, performance and compatibility. We find that safety standards affect adoption through a market access mechanism, perceived performance of risk standards positively impacts customers’ perceived quality, and compatibility standards influence consumers’ perceived network value. The perceived risk, perceived quality and perceived network value influence consumer adoption willingness and behavior. The study contributes to the theory of innovation diffusion and consumer adoption behavior, and offers insights for standardizing activity, innovation diffusion and marketing product information for electric vehicles.

Cross-modal hashing is an important approach for multimodal data management and application. Existing unsupervised cross-modal hashing algorithms mainly rely on data features in pre-trained models to mine their similarity relationships. However, their optimization objectives are based on the static metric between the original uni-modal features, without further exploring data correlations during the training. In addition, most of them mainly focus on association mining and alignment among pairwise instances in continuous space but ignore the latent structural correlations contained in the semantic hashing space. In this paper, we propose an unsupervised hash learning framework ASSPH to solve the above problems. Firstly, we propose an adaptive learning scheme, with limited data and training batches, to enrich semantic correlations of unlabeled instances during the training process and meanwhile to ensure a smooth convergence of the training process. Secondly, we present an asymmetric structural semantic representation learning scheme. We introduce structural semantic metrics based on graph adjacency relations and meanwhile align the inter- and intra-modal semantics in the hash space with an asymmetric binary optimization process. Finally, we conduct extensive experiments to validate the enhancements of our work in comparison with existing works.