Multimodal graphs, where nodes contain heterogeneous features such as images and text, are increasingly common in real-world applications. Effectively learning on such graphs requires both adaptive intra-modal message passing and efficient inter-modal aggregation. However, most existing approaches to multimodal graph learning are typically extended from conventional graph neural networks and rely on static structures or dense attention, which limit flexibility and expressive node embedding learning. In this paper, we propose a novel multimodal graph representation learning framework with Dynamic information Pathways (DiP). By introducing modality-specific pseudo nodes, DiP enables dynamic message routing within each modality via proximity-guided pseudo-node interactions and captures inter-modality dependence through efficient information pathways in a shared state space. This design achieves adaptive, expressive, and sparse message propagation across modalities with linear complexity. We conduct the link prediction and node classification tasks to evaluate performance and carry out full experimental analyses. Extensive experiments across multiple benchmarks demonstrate that DiP consistently outperforms baselines.

Multivariate motif discovery aims to identify frequently occurring subsequences within multi-dimensional time series, which is a critical machine learning task with wide applications. However, previous motif discovery algorithms often miss complex multivariate motifs and struggle with high computational costs as data scale and dimensionality grow. We propose a novel learnable multivariate matrix profile method (L-MAP) that captures inter-dimensional dependencies for comprehensive analysis of multivariate time series. The time series is partitioned into subsequences using the Fourier transform in the frequency domain, with locality-sensitive hashing (LSH) assigning them to buckets based on distinct patterns. Each subsequence is modeled as a graph for multivariate fusion, where triplet learning is used to capture cross-dimensional relationships and form graph embeddings. Unlike prior methods relying on Euclidean distance modeling, our graph-based approach computes all-pairs similarity in a latent space, which constructs the multivariate matrix profile from distributions formed by embedding clusters. Extensive experiments on multivariate datasets from diverse domains demonstrate that L-MAP outperforms state-of-the-art methods in motif discovery, offering superior quality, diversity, and scalability efficiency.

The rapid growth of web content has spurred the widespread adoption of on-device AI assistants powered by large language models (LLMs). However, deploying and personalizing these assistants in real-world environments remains challenging due to limited annotation budgets and scarce on-device fine-tuning resources. Existing edge–cloud collaboration frameworks typically rely on costly cloud-based supervision or perform full-layer finetuning, leading to inefficiencies in both computation and adaptation. To address these limitations, we propose EcoTune, a budget-constrained framework for efficient edge–cloud collaborative adaptation. EcoTune jointly optimizes representative data selection for cloud annotation and selective on-device model adaptation within a unified closed-loop process. Specifically, it employs a multi-armed bandit–based strategy to identify highvalue user interactions for cloud supervision and a layer importance–driven adaptation mechanism to update only critical components of the small language model (SLM). This coordinated optimization enables dynamic, resource-efficient personalization under stringent annotation and tuning budgets. Experiments on real-world testbeds demonstrate that EcoTune achieves up to 20%-60% reduction in annotation costs and significantly lowers fine-tuning memory consumption compared to state-of-the-art baselines, providing a practical and scalable solution for personalized on-device LLMs.

Time series analysis is a fundamental data mining task that has made encouraging progress in many real-world scenarios. Supervised training methods based on empirical risk minimization have proven their effectiveness on specific tasks and datasets. However, the acquisition of well-annotated data is costly, and a large amount of unlabeled series data is under-utilized. Due to distributional shifts across various domains and different patterns of interest across multiple tasks. The problem of cross-domain multi-task migration remains a significant challenge. To address these problems, this article proposes a novel cross-domain approach based on Wave Quantization for Time Series (termed as WQ4TS), which can be combined with any advanced time series model and applied to diverse downstream tasks. Specifically, we transfer the data from different domains into a common spectral latent space and enable the model to learn the temporal pattern knowledge of different domains directly from the common space and utilize it for the inference of downstream tasks, thereby mitigating the challenge of heterogeneous migration. The establishment of spectral latent space brings at least three benefits, cross-domain migration capability thus adapting to zero- and few-shot scenarios without relying on priori knowledge, general compatible cross-domain framework without changing the existing model structure, and robust modeling capability thus achieving SOTA results in multiple downstream tasks. To demonstrate the effectiveness of the proposed approach, we conduct extensive experiments including three important tasks: forecasting, imputation, and classification. And three common real-world scenarios are simulated: full-data, few-shot, and zero-shot. The proposed WQ4TS achieves the best performance on 87.5% of all tasks. Concretely, WQ4TS achieved 25.8% and 44.1% improvements in MSE metric for few-shot and zero-shot forecasting tasks, respectively, and demonstrated excellent 24.9% increase in average accuracy on few-shot classification tasks. The source codes of WQ4TS are publicly available on https://github.com/Xiang-Kai/WQ4TS .

Partial multi-view classification (PMvC) poses a significant challenge due to the incomplete nature of multi-view data, which complicates effective information fusion and accurate classification. Existing PMvC methods typically rely on heuristic evaluations of view informativeness to achieve global alignment for downstream classification tasks. However, these approaches suffer from two critical issues: information redundancy and semantic misalignment. The complexity of missing data not only leads to over-reliance on redundant or less informative views but also exacerbates semantic misalignment across views, making it difficult for existing methods to effectively capture and discriminate the class-related features. To address these issues, this work proposes a novel GLobal-semantic Alignment Distillation (GLAD) model for partial multi-view classification without requiring imputation. Our approach incorporates a self-distillation mechanism that enables the model to extract informative features and achieve global semantic alignment across views. The key insight of GLAD is leveraging labels as semantic anchors to guide the alignment of partial multi-view features. By integrating labels with extracted features via a cross-attention mechanism, we generate ideal embeddings that consistently capture global semantics across views. These embeddings then serve as intermediate supervision for distilling the student model, ensuring robust semantic alignment even with missing views. We further introduce a margin-aware weighting strategy to enhance the model's discriminative ability. Extensive experimental results validate the effectiveness and superiority of the proposed method, showcasing significant improvements in classification performance over existing techniques.

Graph Neural Networks (GNNs) have proven effective and typically benefit from pre-training on accessible graphs to enhance performance on tasks with limited labeled data. However, existing GNNs are constrained by the ``one-domain-one-model'' limitation, which restricts their effectiveness across diverse graph domains. In this paper, we tackle this problem by developing a method called Multi-Domain Pre-training for a Unified GNN Model (MDP-GNN). This method is based on the philosophical notion that everything is interconnected, suggesting that a latent meta-domain exists to encompass the diverse graph domains and their interconnections. MDP-GNN seeks to identify and utilize this meta-domain to train a unified GNN model through three core strategies. Firstly, it integrates node feature semantics from different domains to create unified representations. Secondly, it employs a bi-level learning strategy to build a domain-synthesized network that identifies latent connections to facilitate cross-domain knowledge transfer. Thirdly, it uses Wasserstein distance to map diverse domains into the common meta-domain for graph distribution alignment. We validate the effectiveness of MDP-GNN through theoretical analysis and extensive experiments on four real-world graph datasets, showing its superiority in enhancing GNN performance across diverse domains.

Graph transfer learning endeavors to develop a Graph Neural Network (GNN) model in a fully-labeled source domain, with the intention of deploying it on a target domain that has limited labeled data for inference. We reveal that prevalent graph transfer learning methods are susceptible to the homophily shift problem. This issue arises from the divergence in homophily structures between the source and target graphs, leading to a notable deterioration in the performance of GNN models. In this paper, we introduce a novel Contextual Structural Graph Neural Network (CS-GNN) method, leveraging a tailored attention mechanism to apprehend a variety of local structural cues, facilitating structural knowledge transfer across domains. It features an ego-network module to distill local structural diversity and a moment-based approach to gauge structural patterns without needing ground-truth labels. CS-GNN crafts a feature smoothness matrix from node attributes, guiding a customized attention mechanism for feature aggregation. A group-wise fairness loss is employed to balance learning across various structural patterns, enhancing the model's ability to transfer knowledge across domains. Comprehensive experiments conducted on six benchmark datasets substantiate the superiority of CS-GNN over the state-of-the-art methods, demonstrating significant improvements in accuracy and robustness against homophily shifts.

LiDAR-based 3D object detection presents significant challenges due to the inherent sparsity of LiDAR points. A common solution involves long-term temporal LiDAR data to densify the inputs. However, efficiently leveraging spatial-temporal information remains an open problem. In this paper, we propose a novel Semantic-Supervised Spatial-Temporal Fusion (ST-Fusion) method, which introduces a novel fusion module to relieve the spatial misalignment caused by the object motion over time and a feature-level semantic supervision to sufficiently unlock the capacity of the proposed fusion module. Specifically, the ST- Fusion consists of a Spatial Aggregation (SA) module and a Temporal Merging (TM) module. The SA module employs a convolutional layer with progressively expanding receptive fields to aggregate the object features from the local regions to alleviate the spatial misalignment, the TM module dynamically extracts object features from the preceding frames based on the attention mechanism for a comprehensive sequential presentation. Besides, in the semantic supervision, we propose a Semantic Injection method to enrich the sparse LiDAR data via injecting the point-wise semantic labels, using it for training a teacher model and providing a reconstruction target at the feature level supervised by the proposed object-aware loss. Extensive experiments on various LiDAR-based detectors demonstrate the effectiveness and universality of our proposal, yielding an improvement of approximately +2.8% in NDS based on the nuScenes benchmark.

Recent Vision-based Large Language Models (VisionLLMs) for autonomous driving have seen rapid advancements. However, such promotion is extremely dependent on large-scale high-quality annotated data, which is costly and labor-intensive. To address this issue, we propose unlocking the value of abundant yet unlabeled data to improve the language-driving model in a semi-supervised learning manner. Specifically, we first introduce a series of template-based prompts to extract scene information, generating questions that create pseudo-answers for the unlabeled data based on a model trained with limited labeled data. Next, we propose a Self-Consistency Refinement method to improve the quality of these pseudo-annotations, which are later used for further training. By utilizing a pre-trained VisionLLM (e.g., InternVL), we build a strong Language Driving Model (LDM) for driving scene question-answering, outperforming previous state-of-theart methods. Extensive experiments on the DriveLM benchmark show that our approach performs well with just 5% labeled data, achieving competitive performance against models trained with full datasets. In particular, our LDM achieves 44.85% performance with limited labeled data, increasing to 54.27 % when using unlabeled data, while models trained with full datasets reach 60.68% on the DriveLM benchmark.

Graph Neural Networks (GNNs) have demonstrated remarkable effectiveness across various tasks but are often hindered by their high computational overhead. GNN-to-MLP distillation provides a promising remedy by transferring knowledge from complex GNNs to lightweight MLPs. However, existing methods largely overlook the differences in aggregation mechanisms and heterogeneous architectures. Simplifying such intricate information into MLP potentially causes information loss or distortion, ultimately resulting in suboptimal performance. This paper proposes an aggregation mechanism enhanced GNN distillation framework (AMEND). AMEND introduces multi-scope aggregation context preservation to replicate the teacher's broad aggregation scopes and an aggregation-enhanced centered kernel alignment method to match the teacher's aggregation patterns. To ensure efficient and robust knowledge transfer, we integrate a manifold mixup strategy, enabling the student to capture the teacher's insights into mixed data distributions. Experimental results on 8 standard and 4 large-scale datasets demonstrate that AMEND consistently outperforms state-of-the-art distillation methods.

Multivariate Time Series Classification (MTSC) is a fundamental data mining task, which is widely applied in the fields like health care and energy management. However, the existing MTSC methods are mostly adapted from univariate versions and model the static patterns among series in the time domain. We argue they fail to capture the inter-dependencies across variables and rarely consider the unique dynamic features in multilevel frequencies, which are susceptible to signal noise and lack sufficient feature extraction capability to achieve satisfactory classification accuracy. To address these issues, we propose a novel framework called Multilevel Dynamic Wavelet Graph Neural Network (MagNet) , which effectively captures inherent temporal-frequency dependencies in multivariate time series data in a global view, facilitating the information flow among inter-related variables and leveraging learnable Graph Neural Networks (GNNs) to uncover dynamic frequency dependencies. We propose an orthogonal temporal convolution layer that utilizes soft orthogonal losses to constrain features learned at different frequency components to reduce feature redundancy. Additionally, we introduce a hierarchical graph coarsening operator to address the flat learning challenges in traditional GNNs. Our dynamic wavelet GNN and hierarchical coarsening enable deep model stacking and end-to-end learning. Extensive experiments on 30 UEA benchmarks demonstrate that our method outperforms the state-of-the-art baselines in the MTSC tasks.

Graph Neural Networks (GNNs) have emerged as a powerful tool for modeling graph-structured data, exhibiting remarkable potential in applications such as social networks, recommendation systems, and molecular structures. However, the conventional GNNs perform node-level feature aggregation from neighbors without considering graph-label information, which leads to the misaligned embedding problem that may cause a detrimental effect on graph-level tasks such as graph classification. In this paper, we propose a novel label-attentive distillation method called LAD-GNN for graph representation learning to solve this problem. It alternatively trains a teacher model and a student GNN with a distillation-based approach. In the teacher model, a label-attentive encoder is proposed to encode the label information fusing with the node features to generate ideal embedding. In the student model, the ideal embedding is used as intermediate supervision to urge the student GNN to learn class-friendly node embedding to facilitate graph-level tasks. Generally, LAD-GNN is an enhanced GNN training approach that can be incorporated with arbitrary GNN backbone to improve performance without significant increase of computational cost. Extensive experiments with 7 GNN backbones based on 10 benchmark datasets show that LAD-GNN improves the SOTA GNNs in graph classification accuracy. The source codes of LAD-GNN are publicly available on https://github.com/XiaobinHong/LAD-GNN.

Time series analysis has a wide range of applications in the fields of weather forecasting, traffic management, fault detection, intelligent operation, etc. In the real world, time series typically consist of dynamic fluctuations and mixtures of periodicities, which bring challenges on modeling and analyzing their patterns. To overcome the complexities, a common approach is to decompose long-term time series into sub-components for easier analysis. Unlike conventional time series decomposition that decouples a series into the trend and seasonal parts, we proposed a novel triple decomposition method to decouple a long-term series into three components: trend-part, regular-part, and fluctuant-part. Notably, the third part is a particular component that represents the dynamic spectral fluctuation in time series with the formulation of spectrum gradient. Based on triple decomposition, we propose a novel task-general deep learning model called TS3Net for long-term series analysis. It introduces a temporal-frequency block (TF-Block) with a multi-branch structure to expand the time series into a 2D temporal-frequency distribution. Subsequently, deep representation can be learned by a vision architecture that captures the dynamic variations from the complex multi-periodic series. The decomposed components are processed by TS3Net individually, and their results are integrated to form the final result for time series analysis. We conduct extensive experiment based on six open datasets to evaluate the proposed method in comparison with 10 baselines. Numerical results show that TS3Net significantly outperforms the state-of-the-art methods on both time series forecasting and imputation tasks. The source codes of TS3Net are publicly available on https://github.com/Xiang-Kai/TS3Net.

Graph Neural Networks (GNNs) have proven effective in various scenarios. A key strategy involves pre-training existing graphs to extract knowledge that can be transferred to improve performance on downstream tasks, reducing the need for extensive labeled data. However, previous works commonly assumed that pre-training and fine-tuning occur in the same or closely related domains. A limitation is that for each individual graph without accessible pre-training data, a GNN must be trained from scratch, imposing high training overhead and hindering the ability of generalization. In this paper, we address the GNN multi-domain pre-training problem, which intends to pre-train a transferable GNN model from heterogeneous multi-source graph domains and then apply it in an unseen one with minor fine-tuning costs. To this end, we propose a scaLA ble Multi-source Pre-training (LAMP) method. For pre-training, LAMP presents a graph dual-distillation approach to distill massive knowledge from various graph domains to form synthetic homogeneous graphs. Simultaneously, high-level meta-knowledge from the synthetic graphs is extracted to train the GNN model, whose capability can be adjusted according to target graph contexts through a co-training modulation architecture. For fine-tuning, LAMP respectively aligns the target graph distribution, graph context, and graph task with the pretext so that the downstream task in the unseen domain can be reshaped to leverage the transferable knowledge efficiently. Extensive experiments on four different graph domain datasets show the superiority of LAMP.

Graph embedding aims to encode nodes/edges into low-dimensional continuous features, and has become a crucial tool for graph analysis including graph/node classification, link prediction, etc. In this paper we propose a novel graph learning framework, named graph game embedding, to learn discriminative node representation as well as encode graph structures. Inspired by the spirit of game learning, node embedding is converted to the selection/searching process of player strategies, where each node corresponds to one player and each edge corresponds to the interaction of two players. Then, a utility function, which theoretically satisfies the Nash Equilibrium, is defined to measure the benefit/loss of players during graph evolution. Furthermore, a collaboration and competition mechanism is introduced to increase the discriminant learning ability. Under this graph game embedding framework, considering different interaction manners of nodes, we propose two specific models, named paired game embedding for paired nodes and group game embedding for group interaction. Comparing with existing graph embedding methods, our algorithm possesses two advantages: (1) the designed utility function ensures the stable graph evolution with theoretical convergence and Nash Equilibrium satisfaction; (2) the introduced collaboration and competition mechanism endows the graph game embedding framework with discriminative feature leaning ability by guiding each node to learn an optimal strategy distinguished from others. We test the proposed method on three public datasets about citation networks, and the experimental results verify the effectiveness of our method.