Knowledge Graph (KG)-supported Graph Neural Network models are becoming crucial in recommendation systems due to their ability to mitigate the data sparsity challenge. However, these models remain suboptimal because they overlook the representation differences between the inherent user-item Bipartite Graph (BG) and the external head-relation-tail KG, leading to semantic misalignment. Moreover, they indiscriminately incorporate various types of relations from the KG, which may introduce noise information into the model, ultimately degrading recommendation performance. To address these challenges, we propose an end-to-end model named Multi-graph Fusion Cross-model Contrastive Learning (MFCCL). To uncover users' interest in items and explore the associations between items, we first construct a user-interest graph by integrating information from both the BG and KG, and an item-association graph derived from the BG. We devise a multi-graph representation learning module that incorporates rich semantics into user and item representations in parallel. Simultaneously, a classical collaborative filtering module is introduced to fully leverage user-item collaborative signals. Additionally, we design a novel free data-augmentation cross-model contrastive learning to facilitate the exchange of complementary information between different models. Empirical evaluations on three widely used benchmarks demonstrate that our MFCCL method achieved significant improvements over the baselines.

Bias in Large Language Models (LLMs) is increasingly addressed through fairness-oriented techniques. However, in some cases, these approaches may inadvertently remove genuine cultural differences between groups, leading to “over-normalization” or models losing important socio-cultural distinctions. In this work, we introduce OverNormEval, a benchmark designed to detect when an LLM exhibits such over-normalization. We further explore the use of Direct Preference Optimization (DPO) to mitigate over-normalization.

Multi-graph multi-label learning (MGML) represents each object as a bag-of-graphs with multiple labels, but demands large-scale labeled data whose acquisition is often difficult and costly. Self-supervised contrastive learning (SCL) mitigates label dependence by leveraging data augmentation to construct discriminative pretext tasks, proving effective for multi-instance learning. However, when applied to MGML, SCL faces two key challenges: (1) it distinguishes individual instances by their differences, whereas MGML requires modeling label correlations; (2) it assumes semantic invariance under augmentation, but structural perturbations in MGML alter label semantics. To tackle these challenges, we propose a self-suPervised contrastive rE-learning framework for mulTi-grAph multi-labeL classification (PETAL). Specifically, to model label correlations, we first define a unified label space to learn label prototypes and align features with them, yielding prototype-aligned representations. We then design a multi-granularity contrastive loss over these representations, which captures label dependencies by contrasting at the bag level, graph level, and bag-graph level. Moreover, to ensure semantic invariance, we develop a contrastive re-learning strategy based on prototype-aligned representations to generate augmentation-free positive samples. This guarantees consistent multi-label distributions without structural perturbations. Experiments on six datasets demonstrate that PETAL achieves an average improvement of 4.12% over state-of-the-art self-supervised and supervised baselines.

Navya Gupta, Rishitej Reddy Vyalla, Avinash Anand, Chhavi Kirtani, Erik Cambria, Zhengchen Zhang, Zhengkui Wang, Timothy Liu, Aik Beng Ng, Simon See, Rajiv Ratn Shah. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Every graph hides a tree: through tree decomposition—a foundational tool in modern graph theory with broad applications such as in computational power networks, any network can be unfolded into a hierarchy of overlapping vertex bags whose backbone is a tree. Leveraging this powerful lens, we propose Topological Decomposition for Self-supervised Learning (TopDSL), a framework that injects multi-scale signals into graph representation learning. Concretely, we: 1) decompose the input graph into tree structures with bags representing local structural contexts; 2) compute bag-level roles via closeness centrality for nodes and local edge betweenness for edges, and aggregate these scores across bags to capture context-dependent importance (e.g., local structural bridges); 3) convert the resulting importance and attribute-stability scores into a context-aware augmentation policy that adaptively perturbs nodes, edges, and features—preserving local bridges, honoring multi-community vertices, and attenuating noisy global hubs; 4) construct a new structural similarity loss for contrastive learning, which fuses traditional graph-based proximity with a novel tree-based similarity derived from node co-occurrence in decomposition bags; 5) demonstrate that our framework achieves superior performance over state-of-the-art baselines on various graph learning benchmarks.

Multi-Graph Learning (MGL) is a fundamental machine learning paradigm that represents objects as bags-of-graphs, each encoding a distinct structural property, and has broad applications in bioinformatics, chemistry, computing power networks, and software defect detection. However, the inherent scarcity of labeled data poses a significant bottleneck for supervised MGL approaches. While self-supervised contrastive learning offers a compelling solution, its direct application to MGL faces three key challenges: (1) existing graph neural networks, primarily for single-graph modeling, struggle to yield discriminative bag-level representations from bags-of-graphs; (2) conventional contrastive objectives are limited to single-level settings, failing to capture cross-hierarchical dependencies; and (3) standard data augmentation often disrupts intrinsic graph and bag structures, undermining semantic consistency. To address these issues, we propose the Hierarchical Graph-Bag-Network (HGBN), a self-supervised MGL framework that constructs hierarchical representations in the form of a graph-bag-network. HGBN employs an asymmetric hierarchical graph neural network to learn discriminative graph-level and bag-level representations, introduces cross-hierarchical contrastive objectives to align graph-level and bag-level semantics, and leverages the asymmetric network outputs to form positive and negative pairs, preserving intrinsic structural and semantic consistency. Experiments on eight benchmark multi-graph datasets demonstrate that HGBN consistently outperforms both supervised and self-supervised state-of-the-art baselines, achieving average improvements of 4.82% in accuracy and F1 score.

Solving partial differential equations (PDEs) for fluid simulation is computationally expensive, especially when dealing with complex geometries and high-resolution meshes. Recent advances in physics-informed graph neural networks (PIGNNs) have demonstrated potential in approximating such simulations more efficiently. Particularly, thermodynamic informed graph neural networks (TIGNNs) offer a promising data-driven alternative to traditional PDE solvers for fluid simulations. However, existing TIGNN implementations suffer from significant training inefficiencies, requiring prolonged runtimes and high memory consumption due to the need to maintain large parameter matrices in GPU memory. Inspired by multi-processor strategies in deformable solid simulations, we propose a novel multi-processor thermodynamic-informed graph neural network (MP-TIGNN) architecture to significantly accelerate training without compromising accuracy. Our approach enables faster convergence and reduces memory usage with mixed-precision training by leveraging fully sharded data parallelism (FSDP) across multiple GPUs. Experimental results show that our approach reduces training time by approximately 70% compared to the original setup while maintaining similar prediction accuracy.

Network resilience refers to the ability of a network to maintain its functionality despite perturbations, where resilience/robustness is shown when a substantial proportion of its nodes remain engaged even under changes. Such phenomenon is common in real-world networks, such as computing power networks. Previous works demonstrate that the coreness of a user/node effectively captures the dynamics of user engagement. However, most existing works only consider changes in a single coreness value and thus fail to measure the overall network resilience. Subsequent works are either inefficient or do not consider coreness-decreased scenario. In this paper, we propose and study the collapsed follower maximization problem, aiming to maximize the number of coreness-decreased vertices by finding a group collapser (collapsing a set of vertices) with a given budget. We prove that the problem is NP-hard and W[2]-hard parameterized by the budget b. To address the problem, we first present a Greedy algorithm that iteratively finds the best collapser in each of the budget b iterations. To further optimize the Greedy algorithm, we propose GreedyOpt, which leverages the shell component structure to accelerate the computation of follower for one collapser and prune the search space. Extensive experimental results on 8 real-world datasets show that the effectiveness and efficiency of our algorithms.

Graph Neural Networks (GNNs) are the rising standard for graph tasks, yet their distributed training in servers or computing power network remains challenging. Cross-machine sampling and data loading often create bottlenecks, leading to inefficient resource utilization. In this paper, we present Hybrid2, a distributed GNN training system that combines full-graph and mini-batch training through a novel hybrid-batch training method. It also adopts hybrid feature extraction, leveraging both local caching and remote access to improve feature retrieval efficiency. The integration of these methods in Hybrid² results in a dual hybrid-gain effect. First, it reduces sampling and loading overhead by pre-aggregating neighbors for each target vertex, minimizing the layers to sample and load. Second, it accelerates data loading by dynamically identifying and locally caching the most frequently accessed vertices during training, maximizing memory efficiency. Experimental results demonstrate that Hybrid² brings substantial performance improvements across key components of distributed GNN training. Network communication overhead is reduced by up to tens of times, while both sampling and loading achieve at least several-fold speedups. These gains contribute to an overall training acceleration exceeding 20× compared to DistDGL, all with comparable GPU memory usage and no loss in accuracy. Compared to the state-of-the-art system, it achieves nearly 3× speedup while using fewer resources.

Node equivalence is common in graphs, such as computing networks, encompassing automorphic equivalence (preserving adjacency under node permutations) and attribute equivalence (nodes with identical attributes). Despite their importance for learning node representations, these equivalences are largely ignored by existing graph models. To bridge this gap, we propose a GrAph self-supervised Learning framework with Equivalence (GALE) and analyze its connections to existing techniques. Specifically, we: 1) unify automorphic and attribute equivalence into a single equivalence class; 2) enforce the equivalence principle to make representations within the same class more similar while separating those across classes; 3) introduce approximate equivalence classes with linear time complexity to address the NP-hardness of exact automorphism detection and handle node-feature variation; 4) analyze existing graph encoders, noting limitations in message passing neural networks and graph transformers regarding equivalence constraints; 5) show that graph contrastive learning are a degenerate form of equivalence constraint; and 6) demonstrate that GALE achieves superior performance over baselines.

Graphs, fundamental in modeling various research subjects such as computing networks, consist of nodes linked by edges. However, they typically function as components within larger structures in real-world scenarios, such as in protein-protein interactions where each protein is a graph in a larger network. This study delves into the Graph-of-Net (GON), a structure that extends the concept of traditional graphs by representing each node as a graph itself. It provides a multi-level perspective on the relationships between objects, encapsulating both the detailed structure of individual nodes and the broader network of dependencies. To learn node representations within the GON, we propose a position-aware neural network for Graph-of-Net which processes both intra-graph and inter-graph connections and incorporates additional data like node labels. Our model employs dual encoders and graph constructors to build and refine a constraint network, where nodes are adaptively arranged based on their positions, as determined by the network’s constraint system. Our model demonstrates significant improvements over baselines in empirical evaluations on various datasets.

To be still yet still moving. - Do Hyun Choe

Graph self-supervised learning aims to learn the intrinsic graph representations from unlabeled data, with broad applicability in areas such as computing networks. Although graph contrastive learning (GCL) has achieved remarkable progress by generating perturbed views via data augmentation and optimizing sample similarity, it performs poorly in heterophilic graph scenarios (where connected nodes are likely to belong to different classes or exhibit dissimilar features). In heterophilic graphs, existing methods typically rely on random or carefully designed augmentation strategies (e.g., edge dropping) for contrastive views. However, such graph structures exhibit intricate edge relationships, where topological perturbations may completely alter the semantics of neighborhoods. Moreover, most methods focus solely on local contrastive signals while neglecting global structural constraints. To address these limitations, inspired by graph coloring, we propose a novel Coloring learning for heterophilic graph Representation framework, CoRep, which: 1) Pioneers a coloring classifier to generate coloring labels, explicitly minimizing the discrepancy between homophilic nodes while maximizing that of heterophilic nodes. A global positive sample set is constructed using multi-hop same-color nodes to capture global semantic consistency. 2) Introduces a learnable edge evaluator to guide the coloring learning dynamically and utilizes the edges' triplet relations to enhance its robustness. 3) Leverages Gumbel-Softmax to differentially discretize color distributions, suppressing noise via a redundancy constraint and enhancing intra-class compactness. Experimental results on 14 benchmark datasets demonstrate that CoRep significantly outperforms current state-of-the-art methods.

Nutrition estimation is crucial for effective dietary management and overall health and well-being. Existing methods often struggle with sub-optimal accuracy and can be time-consuming. In this paper, we propose NuNet, a transformer-basednetwork designed fornutrition estimation that utilizes both RGB and depth information from food images. We have designed and implemented a multi-scale encoder and decoder, along with two types of feature fusion modules, specialized for estimating five nutritional factors. These modules effectively balance the efficiency and effectiveness of feature extraction with flexible usage of our customized attention mechanisms and fusion strategies. Our experimental study shows that NuNet significantly outperforms its variants and existing solutions for nutrition estimation. It achieves an error rate of 15.65%, the lowest known to us, largely due to our multi-scale architecture and fusion modules. This research holds practical values for dietary management with huge potential for transnational research and deployment and could inspire other applications involving multiple data types with varying degrees of importance.

Limited-supervised multi-label learning (LML) leverages weak or noisy supervision for multi-label classification model training over data with label noise, which contain missing labels and/or redundant labels. Existing studies usually solve LML problems by assuming that label noise is independent of the input features and class labels, while ignoring the fact that noisy labels may depend on the input features (instance-dependent) and the classes (label-dependent) in many real-world applications. In this paper, we propose limited-supervised Multi-label Learning with Dependency Noise (MLDN) to simultaneously identify the instance-dependent and label-dependent label noise by factorizing the noise matrix as the outputs of a mapping from the feature and label representations. Meanwhile, we regularize the problem with the manifold constraint on noise matrix to preserve local relationships and uncover the manifold structure. Theoretically, we bound noise recover error for the resulting problem. We solve the problem by using a first-order scheme based on proximal operator, and the convergence rate of it is at least sub-linear. Extensive experiments conducted on various datasets demonstrate the superiority of our proposed method.

Attributed Heterogeneous Information Networks (AHINs) amalgamate the advantages of attributed graphs (AGs) and heterogeneous information networks (HINs) to model intri-cate systems. Within this context, community search-aiming to identify the most probable community containing the queried ver-tex-has been extensively explored in AGs and HINs. However, existing methodologies fall short in simultaneously accommodating heterogeneous attributes and multiple meta-paths in AHINs, posing a substantial challenge in investigating community search within expansive AHINs. Recent studies highlight the efficacy of machine learning-based community search, offering enhanced flexibility and higher-quality communities in comparison to traditional structural-based methods. Yet, semi-supervised learning methods demand substantial labeled data and incur considerable memory and time costs when applied to large AHINs. To tackle these challenges, we propose a MK (Most-likely; K-sized) community search approach. This approach involves defining an MK community and leveraging Graph Neural Networks (GNNs) to amalgamate structures and attributes into a unified goodness metric. Our methodology involves training on local subgraphs sampled via guided random walks based on multiple meta-paths, circumventing the need for training on the entire graph. Moreover, attention-based GNNs adeptly learn meta-path weights to guide weighted walks in subsequent iterations. Additionally, self-training is employed to alleviate the labeling burden. We also demonstrate that pinpointing the location for the MK community is NP-hard and present a heuristic local search strategy that expedites the resolution process through rewriting. Ultimately, the convergence of iterations yields the solution. Extensive experiments conducted on four real-world datasets underscore that the MK framework significantly enhances both effectiveness and efficiency in community search within AHINs. Our code is publicly available at https://github.com/uucxuu/CSAH.

Community search is a fundamental problem in graph analysis. However, prevailing community search models predominantly focus on non-attributed or vertex-attributed graphs. Real-world graphs often bear crucial information within their edges, depicting intricate interactions among vertices. Integrating this edge-based information becomes pivotal in refining community search methodologies. In this paper, we proposed the Edge-Attributed Community Search (EACS) problem and proved that the EACS problem is NP-hard. Advanced exact and 2-approximation algorithms are proposed to address the EACS problem. Extensive experiments demonstrate the efficiency and effectiveness of our algorithms.

Multi-label text classification (MLTC) holds significant importance in the field of data management and information retrieval, where the distribution of label samples often exhibits a long-tail pattern. This poses a formidable challenge for effectively summarizing test data, particularly for classes with limited sample representation. Existing approaches aimed at addressing this challenge often involve altering the original data distribution, resulting in reduced generalization performance on real-world data and inadequate mitigation of the long-tail problem in samples. To tackle this challenge, we propose a novel method for multi-label text classification called MLTC-LD (an enhanced Multi-Label Text Classification model by incorporating Label Dependency). The primary objective of MLTC-LD is to enhance the performance of classes with scarce data by leveraging knowledge acquired from classes with abundant data. MLTC-LD leverages Graph Attention Networks (GAT) to incorporate prior information associated with labels. It calculates a relationship matrix between label samples and integrates this information into the classifier. We conducted a comprehensive evaluation of MLTC-LD on three datasets and compared its performance against eight baseline models. The experimental results validate the superiority of MLTC-LD in effectively mitigating the challenges posed by long-tail distributions.

This paper proposes SpeeDF, a novel three-step framework for anonymizing speech data, particularly focusing on Singaporean English (Singlish). SpeeDF tackles the challenge of protecting less-studied Personally Identifiable Information (PII) like NRIC and passport numbers, which often go overlooked by traditional de-identification methods. Unlike approaches focused solely on entity extraction, SpeeDF leverages a combination of automatic speech recognition (ASR), named entity recognition (NER), and information anonymization. This comprehensive approach ensures thorough PII redaction while preserving the naturalness and usability of the anonymized speech data for research and various downstream applications.

Abstract Pedestrian detection is crucial for various applications, including intelligent transportation and video surveillance systems. Although recent research has advanced pedestrian detection models like the YOLO series, they still face limitations in handling diverse pedestrian scales, leading to performance challenges. To address these issues, we propose HF-YOLO, an advanced pedestrian detection model. HF-YOLO tackles the complexities of pedestrian detection in complex scenes by addressing scale variations and occlusions among pedestrians. In the feature fusion stage, our algorithm leverages both shallow localization information and deep semantic information. This involves fusing P2 layer features and adding a high-resolution detection layer, significantly improving the detection of small-scale pedestrians and occluded instances. To enhance feature representation, HF-YOLO incorporates the HardSwish activation function, introducing more non-linear factors and strengthening the model’s ability to represent complex and discriminative features. Additionally, to address regression imbalance, a balance factor is introduced to the CIoU loss function. This modification effectively resolves the imbalance problem and enhances pedestrian localization accuracy. Experimental results demonstrate the effectiveness of our proposed algorithm. HF-YOLO achieves notable improvements, including a 3.52% increase in average precision, a 1.35% boost in accuracy, and a 4.83% enhancement in recall. Moreover, the algorithm maintains real-time performance with a detection time of 8.5ms, meeting the stringent requirements of real-time applications.

Multi-graph Multi-label learning (Mgml) aims to classify a set of objects of interest, such as text or images, using a bag-of-graphs representation. PreviousMgmlworks have limitations as they only learn labels at the bag level, lose structural information in learning by transferring graphs into instances, and cannot handle noisy labels. This paper presents a robustcoarseandfine-grainedNoise Multi-graph Multi-label (cfMGNML) learning framework that builds the learning model over the graphs and empowers label prediction at both thecoarse(bag) andfine-grained(graph in each bag) levels with noisy labels. To identify label noise, a label probability matrix is defined to act on the scoring function of each label, with a higher probability value indicating that the label is more likely to be the corresponding graph or bag label. The problem is regularized with the manifold constraint on the label probability matrix to preserve local relationships within the data and uncover its essential manifold structure. Meanwhile, a thresholding rank-loss objective is proposed to rank the labels for the graphs and bags and minimize the hamming loss at one step simultaneously. To tackle the non-convex optimization problem, an effective sub- gradient descent algorithm is developed. Experiments over various datasets demonstrate the proposed method achieves superior performance than the state-of-the-art algorithms.

Multi-instance learning (MIL) is a supervised learning where each example is a labeled bag with many instances. The typical MIL strategies are to train an instance-level feature extractor followed by aggregating instances features as bag-level representation with labeled information. However, learning such a bag-level representation highly depends on a large number of labeled datasets, which are difficult to get in real-world scenarios. In this paper, we make the first attempt to propose a robust Self-supervised Multi-Instance LEarning architecture with Structure awareness (SMILEs) that learns unsupervised bag representation. Our proposed approach is: 1) permutation invariant to the order of instances in bag; 2) structure-aware to encode the topological structures among the instances; and 3) robust against instances noise or permutation. Specifically, to yield robust MIL model without label information, we augment the multi-instance bag and train the representation encoder to maximize the correspondence between the representations of the same bag in its different augmented forms. Moreover, to capture topological structures from nearby instances in bags, our framework learns optimal graph structures for the bags and these graphs are optimized together with message passing layers and the ordered weighted averaging operator towards contrastive loss. Our main theorem characterizes the permutation invariance of the bag representation. Compared with state-of-the-art supervised MIL baselines, SMILEs achieves average improvement of 4.9%, 4.4% in classification accuracy on 5 benchmark datasets and 20 newsgroups datasets, respectively. In addition, we show that the model is robust to the input corruption.

The worldwide gaming peripheral market is expanding significantly due to the increasing popularity of online games, and it is predicted that this would increase demand for gaming peripherals. Brand recognition is just the start of the process because many sectors are vying to stand out and wrest mindshare away from rivals. In this paper, we presented a tool named BrandTrend, which enables automated insight discovery for game trending, gaming influencers, and gaming product promotion. The data used in this tool is gathered from social media platforms to analyse gaming contents to match gaming content creators with gaming peripheral brands to promote their brand products via social media. Utilizing data analysis and incorporating evidence from data to build predictions and develop strategies can unambiguously address the issue of distinguish oneself from other rivals and get recognition.

COVID-19's impacts have spread widely in all directions such as economy, people's lifestyles and well-being. Though existing studies have highlighted such an impact, it remains unclear how the current COVID-19 situation has affected the retrenchment, vaccination and global happiness. In this paper, we present an automated tool enables the public to view various insight. In particular, we integrate and analyze the data from various data sources and show how the COVID19 has impacted Singapore and globally. We employ the regression models to identify the correlation between Human Development Index, Stringency Index, Gross Domestic Product per Capita, Total Deaths from COVID-19, and Total Cases of COVID-19; the rate of vaccination and vaccine hesitancy; and the factors to positively correlate to the global happiness. The insight provided adds values to better fight against the COVID-19 pandemic and future global crisis.

Community search is a fundamental graph processing task that aims to find a community containing the given query node. Recent studies show that machine learning (ML)-based community search can return higher-quality communities than the classic methods such as k-core and k-truss. However, the state-of-the-art ML-based models require a large number of labeled data (i.e., nodes in ground-truth communities) for training that are difficult to obtain in real applications, and incur unaffordable memory costs or query time for large datasets. To address these issues, in this paper, we present the community search based on contrastive learning with partition, namely COCLEP, which only requires a few labels and is both memory and query efficient. In particular, given a small collection of query nodes and a few (e.g., three) corresponding ground-truth community nodes for each query, COCLEP learns a query-dependent model through the proposed graph neural network and the designed label-aware contrastive learner. The former perceives query node information, low-order neighborhood information, and high-order hypergraph structure information, the latter contrasts low-order intra-view, high-order intra-view, and low-high-order inter-view representations of the nodes. Further, we theoretically prove that COCLEP can be scalable to large datasets with the min-cut over the graph. To the best of our knowledge, this is the first attempt to adopt contrastive learning for community search task that is nontrivial. Extensive experiments on real-world datasets show that COCLEP simultaneously achieves better community effectiveness and comparably high query efficiency while using fewer labels compared with the-state-of-the-art approaches and is scalable for large datasets.

This paper presents a novel spatial query, skyline micro-cluster (SMC) query. Given a set of data points P, a query point q, a radius γ and a density parameter k, the SMC query returns the skyline micro-clusters (MCs), where MC is a set of points in P that can be covered by a circle with radius γ and the number of points in MC is at least k. In this paper, we formally define the SMC query. As the brute-force approach to solving the SMC query in massive datasets has high computation and memory costs, we propose a basic skyline micro-cluster query algorithm, BSMC, which can reduce the time complexity from O(2N) to O(N3). Furthermore, on top of BSMC, we propose an efficient skyline micro-cluster query algorithm (ESMC). In ESMC, we use the z-value index and propose a filter to remove the invalid micro-clusters, which reduces significant computation overhead. To reduce the memory overhead, we propose an incremental skyline query method. A comprehensive performance study is conducted on real datasets and the experimental results show that our proposed method, ESMC, can significantly improve the SMC query performance.

Mixing data augmentation methods have been widely used in text classification recently. However, existing methods do not control the quality of augmented data and have low model explainability. To tackle these issues, this paper proposes an explainable text classification solution based on attentive and targeted mixing data augmentation, ATMIX. Instead of selecting data for augmentation without control, ATMIX focuses on the misclassified training samples as the target for augmentation to better improve the model's capability. Meanwhile, to generate meaningful augmented samples, it adopts a self-attention mechanism to understand the importance of the subsentences in a text, and cut and mix the subsentences between the misclassified and correctly classified samples wisely. Furthermore, it employs a novel dynamic augmented data selection framework based on the loss function gradient to dynamically optimize the augmented samples for model training. In the end, we develop a new model explainability evaluation method based on subsentence attention and conduct extensive evaluations over multiple real-world text datasets. The results indicate that ATMIX is more effective with higher explainability than the typical classification models, hidden-level, and input-level mixup models.

Self-supervised learning on graph aims to learn graph representations in an unsupervised manner. While graph contrastive learning (GCL - relying on graph augmentation for creating perturbation views of anchor graphs and maximizing/minimizing similarity for positive/negative pairs) is a popular self-supervised method, it faces challenges in finding label-invariant augmented graphs and determining the exact extent of similarity between sample pairs to be achieved. In this work, we propose an alternative self-supervised solution that (i) goes beyond the label invariance assumption without distinguishing between positive/negative samples, (ii) can calibrate the encoder for preserving not only the structural information inside the graph, but the matching information between different graphs, (iii) learns isometric embeddings that preserve the distance between graphs, a by-product of our objective. Motivated by optimal transport theory, this scheme relays on an observation that the optimal transport plans between node representations at the output space, which measure the matching probability between two distributions, should be consistent to the plans between the corresponding graphs at the input space. The experimental findings include: (i) The plan alignment strategy significantly outperforms the counterpart using the transport distance; (ii) The proposed model shows superior performance using only node attributes as calibration signals, without relying on edge information; (iii) Our model maintains robust results even under high perturbation rates; (iv) Extensive experiments on various benchmarks validate the effectiveness of the proposed method.

Given a graph, searching for a community containing a query vertex is a fundamental problem and has found many applications. Most existing community search models are based on non-attributed or vertex-attributed graphs. In many real-world graphs, however, the edges carry the richest information to describe the interactions between vertices; hence, it is important to take the information into account in community search. In this paper, we conduct a pioneer study on the community search on edge-attributed graphs. We proposed the Edge-Attributed Community Search (EACS) problem, which aims to extract a subgraph that contains the given query vertex while its edges have the maximum attribute similarity. We prove that the EACS problem is NP-hard and propose both exact and 2-approximation algorithms to address EACS. Our exact algorithms run up to 2320.34 times faster than the baseline solution. Our approximate algorithms further improve the efficiency by up to 2.93 times. We conducted extensive experiments to demonstrate the efficiency and effectiveness of our algorithms.

Object detection is playing an important role in computer vision. With the rapid development of deep learning, object detection algorithms based on convolutional neural networks have been successful. However, deep learning requires a large amount of data to train, it is crucial to improve model efficiency with the same required hardware resources. In this paper, we propose an efficient one-stage object detection model, EfficientFCOS based on pixel-level prediction, which can improve the accuracy and efficiency of existing one-stage object detection network model. In particular, EfficientFCOS first uses EfficientNet to extract the input image features, which has fewer parameters and better performance than ResNet. Next, it employs the scaling approaches to uniformly scale number of channels of the model’s backbone network, feature fusion network and shared head network according to the resolution of the image. Furthermore, it adopts the weighted bidirectional feature pyramid network to achieve fast and multi-scale feature fusion. Meanwhile, it integrates the geometric factors (center point distance, overlap rate and scale) to the regression loss of target prediction, such that the regression of the target prediction box becomes more stable. Our experimental results indicate that EfficientFCOS is more efficient than the existing FCOS. On the Pascal Voc data set, EfficientFCOS achieves 82.1 for the mAP value, which increases 2.8%. It has 15M network parameters, which is 4.3x smaller than FCOS. Meanwhile, it improves the GPU LAT and CPU LAT by 12.2% and 20.0% respectively.

Supervised clustering methods cluster images using graph convolutional networks (GCN) via linkage prediction, and have shown significant improvements over the traditional clustering algorithms (e.g., K-means, DBScan, etc.) in terms of clustering effectiveness. However, existing supervised clustering approaches are always time-consuming, which may limit their usage. The high computation overhead is mainly resulted from generating and processing a large amount of subgraphs, each of which is generated for one image instance in order to infer the linkage between them. To tackle the high computation problem, we propose a new density division clustering approach based on GCN, and our experiments demonstrate that the new approach is both time-efficient and effective. The approach divides the data into high-density and low-density parts, and only performs GCN subgraph link inference on the low-density parts, which highly reduces redundant calculations. Meanwhile, to ensure sufficient contextual information extraction for low-density parts, it generates adaptive subgraphs instead of fixed-size subgraphs. Our experimental evaluations over multiple datasets show that our proposed approach is five-time faster than state-of-the-art algorithms with even higher accuracy.

Semantic code search greatly facilitates software reuse, which enables users to find code snippets highly matching user-specified natural language queries. Due to the rich expressive power of code graphs (e.g., control-flow graph and program dependency graph), both of the two mainstream research works (i.e., multi-modal models and pre-trained models) have attempted to incorporate code graphs for code modelling. However, they still have some limitations: First, there is still much room for improvement in terms of search effectiveness. Second, they have not fully considered the unique features of code graphs.

Density Peaks (DP) Clustering organizes data into clusters by finding peaks in dense regions. This involves computing density ($\rho$) and distance ($\delta$) of every point. As such, though DP has been very effective in producing high quality clusters, their complexity is O($N^2$) where$N$is the number of data points. In this paper, we propose a fast distributed density peaks clustering algorithm, FDDP, based on the z-value index. In FDDP, we first employ the z-value index to map multi-dimensional data points into one dimensional space, and then range-partition the data according to the z-value to balance the load across the processing nodes. We ensure minimal overlapping range to handle computations at the boundary points. We also propose FC, an efficient algorithm that employs a forward computing strategy to calculate$\rho$linearly. Additionally, we propose another algorithm, CB, which uses a caching and efficient searching strategy to compute$\delta$. Moreover, FDDP is able to reduce the time complexity from$O(N^2)$to$O(N\cdot log(N))$. We provide a theoretical analysis of FDDP and evaluated FDDP empirically. Our experimental results show that FDDP outperforms the state-of-the-art algorithms significantly.

Density Peaks (DP) Clustering organizes data into clusters by finding peaks in dense regions. This involves computing density (ρ) and distance (δ) of every point, and the time complexity is O(N2) where N is the number of data points. In this paper, we propose a fast distributed density peaks clustering algorithm, FDDP, based on the z-value index. We also propose FC, an efficient algorithm that employs a forward computing strategy to calculate ρ linearly, and CB, which uses a caching and efficient searching strategy to compute δ. Our experimental results show that FDDP outperforms the state-of-the-art algorithms significantly.

Graphs are a powerful and versatile data structure that easily captures real life relationship. Multi-graph Multi-label learning (MGML) is a supervised learning task, which aims to learn a Multi-label classifier to label a set of objects of interest (e.g. image or text) with a bag-of-graphs representation. However, prior techniques on the MGML are developed based on transferring graphs into instances that does not fully utilize the structure information in the learning, and focus on learning the unseen labels only at the bag level. There is no existing work studying how to label the graphs within a bag that is of importance in many applications like image or text annotation. To bridge this gap, in this paper, we present a novel coarse and fine-grained Multi-graph Multi-label (cfMGML) learning framework which directly builds the learning model over the graphs and empowers the label prediction at both the coarse (aka. bag) level and fine-grained (aka. graph in each bag) level. In particular, given a set of labeled multi-graph bags, we design the scoring functions at both graph and bag levels to model the relevance between the label and data using specific graph kernels. Meanwhile, we propose a thresholding rank-loss objective function to rank the labels for the graphs and bags and minimize the hamming-loss simultaneously at one-step, which aims to address the error accumulation issue in traditional rank-loss algorithms. To tackle the non-convex optimization problem, we further develop an effective sub-gradient descent algorithm to handle high-dimensional space computation required in cfMGML. Experiments over various real-world datasets demonstrate cfMGML achieves superior performance than the state-of-arts algorithms.

Captioning the images with proper descriptions automatically has become an interesting and challenging problem. In this paper, we present one joint model AICRL, which is able to conduct the automatic image captioning based on ResNet50 and LSTM with soft attention. AICRL consists of one encoder and one decoder. The encoder adopts ResNet50 based on the convolutional neural network, which creates an extensive representation of the given image by embedding it into a fixed length vector. The decoder is designed with LSTM, a recurrent neural network and a soft attention mechanism, to selectively focus the attention over certain parts of an image to predict the next sentence. We have trained AICRL over a big dataset MS COCO 2014 to maximize the likelihood of the target description sentence given the training images and evaluated it in various metrics like BLEU, METEROR, and CIDEr. Our experimental results indicate that AICRL is effective in generating captions for the images.

This paper proposed an automated staff roster planning system (SRPS) for the healthcare sector with the objective to revolutionize the existing approach of manual staff roster planning by significantly reducing the number of man-hours used in the process of planning and minimizing the possibility of human error involved in the process. SRPS leverages on its subsystems, the staff training record management system (TRMS) and the leave projection system (LPS) to retrieve and process the competency and availability of staff that are recorded earlier. It generates the feasible staff roster based on the existing available staff, their skillsets and the required skillsets of the workstations. SRPS also takes soft constraints into consideration during the staff to workstation allocation process. The key functionalities of the system are automated staff roster generation, pattern scheduling, constraint scheduling, Gale Shapley matching algorithm, drag and drop roster editing, interactive workstation compatibility feedback and on click export. Evaluation of SRPS shows improvement in terms of time efficiency and allocation optimization.

Shared-nothing architecture has been widely used in distributed databases to achieve good scalability. While it offers superior performance for local transactions, the overhead of processing distributed transactions can degrade the system performance significantly. The key contributor to the degradation is the expensive two-phase commit (2PC) protocol used to ensure atomic commitment of distributed transactions. In this paper, we propose a transaction management scheme called LEAP to avoid the 2PC protocol within distributed transaction processing. Instead of processing a distributed transaction across multiple nodes, LEAP converts the distributed transaction into a local transaction. This benefits the processing locality and facilitates adaptive data repartitioning when there is a change in data access pattern. Based on LEAP, we develop an online transaction processing (OLTP) system, L-Store, and compare it with the state-of-the-art distributed in-memory OLTP system, H-Store, which relies on the 2PC protocol for distributed transaction processing, and H^L-Store, a H-Store that has been modified to make use of LEAP. Results of an extensive experimental evaluation show that our LEAP-based engines are superior over H-Store by a wide margin, especially for workloads that exhibit locality-based data accesses.

Efficient and scalable stream joins play an important role in performing real-time analytics for many cloud applications. However, like in conventional database processing, online theta-joins over data streams are computationally expensive and moreover, being memory-based processing, they impose high memory requirement on the system. In this paper, we propose a novel stream join model, called join-biclique, which organizes a large cluster as a complete bipartite graph. Join-biclique has several strengths over state-of-the-art techniques, including memory-efficiency, elasticity and scalability. These features are essential for building efficient and scalable streaming systems. Based on join-biclique, we develop a scalable distributed stream join system, BiStream, over a large-scale commodity cluster. Specifically, BiStream is designed to support efficient full-history joins, window-based joins and online data aggregation. BiStream also supports adaptive resource management to dynamically scale out and down the system according to its application workloads. We provide both theoretical cost analysis and extensive experimental evaluations to evaluate the efficiency, elasticity and scalability of BiStream.

Attributed graphs are becoming important tools for modeling information networks, such as the Web and various social networks (e.g. Facebook, LinkedIn, Twitter). However, it is computationally challenging to manage and analyze attributed graphs to support effective decision making. In this paper, we propose, Pagrol, a parallel graph OLAP (Online Analytical Processing) system over attributed graphs. In particular, Pagrol introduces a new conceptual Hyper Graph Cube model (which is an attributed-graph analogue of the data cube model for relational DBMS) to aggregate attributed graphs at different granularities and levels. The proposed model supports different queries as well as a new set of graph OLAP Roll-Up/Drill-Down operations. Furthermore, on the basis of Hyper Graph Cube, Pagrol provides an efficient MapReduce-based parallel graph cubing algorithm, MRGraph-Cubing, to compute the graph cube for an attributed graph. Pagrol employs numerous optimization techniques: (a) a self-contained join strategy to minimize I/O cost; (b) a scheme that groups cuboids into batches so as to minimize redundant computations; (c) a cost-based scheme to allocate the batches into bags (each with a small number of batches); and (d) an efficient scheme to process a bag using a single MapReduce job. Results of extensive experimental studies using both real Facebook and synthetic datasets on a 128-node cluster show that Pagrol is effective, efficient and scalable.

In many scientific applications, it is critical to determine if there is a relationship between a combination of objects. The strength of such an association is typically computed using some statistical measures. In order not to miss any important associations, it is not uncommon to exhaustively enumerate all possible combinations of a certain size. However, discovering significant associations among hundreds of thousands or even millions of objects is a computationally intensive job that typically takes days, if not weeks, to complete. We are, therefore, motivated to provide efficient and practical techniques to speed up the processing exploiting parallelism. In this paper, we propose a framework, COSAC, for such combinatorial statistical analysis for large-scale data sets over a MapReduce-based cloud computing platform. COSAC operates in two key phases: 1) In the distribution phase, a novel load balancing scheme distributes the combination enumeration tasks across the processing units; 2) In the statistical analysis phase, each unit optimizes the processing of the allocated combinations by salvaging computations that can be reused. COSAC also supports a more practical scenario, where only a selected subset of objects need to be analyzed against all the objects. As a representative application, we developed COSAC to find combinations of Single Nucleotide Polymorphisms (SNPs) that may interact to cause diseases. We have evaluated our framework on a cluster of more than 40 nodes. The experimental results show that our framework is computationally practical, efficient, scalable, and flexible.

MOTIVATION: Recent studies suggested that a combination of multiple single nucleotide polymorphisms (SNPs) could have more significant associations with a specific phenotype. However, to discover epistasis, the epistatic interactions of SNPs, in a large number of SNPs, is a computationally challenging task. We are, therefore, motivated to develop efficient and effective solutions for identifying epistatic interactions of SNPs. RESULTS: In this article, we propose an efficient Cloud-based Epistasis cOmputing (eCEO) model for large-scale epistatic interaction in genome-wide association study (GWAS). Given a large number of combinations of SNPs, our eCEO model is able to distribute them to balance the load across the processing nodes. Moreover, our eCEO model can efficiently process each combination of SNPs to determine the significance of its association with the phenotype. We have implemented and evaluated our eCEO model on our own cluster of more than 40 nodes. The experiment results demonstrate that the eCEO model is computationally efficient, flexible, scalable and practical. In addition, we have also deployed our eCEO model on the Amazon Elastic Compute Cloud. Our study further confirms its efficiency and ease of use in a public cloud. AVAILABILITY: The source code of eCEO is available at http://www.comp.nus.edu.sg/~wangzk/eCEO.html. CONTACT: [email protected].

The 1000 Genome project has made available a large number of single nucleotide polymorphisms (SNPs) for genome-wide association studies (GWAS). However, the large number of SNPs has also rendered the discovery of epistatic interactions of SNPs computationally expensive. Parallelizing the computation offers a promising solution. In this paper, we propose a cloud-based epistasis computing (CEO) model that examines all k-locus SNPs combinations to find statistically significant epistatic interactions efficiently. Our CEO model uses the MapReduce framework which can be executed both on user's own clusters or on a cloud environment. Our cloud-based solution offers elastic computing resources to users, and more importantly, makes our approach affordable and available to all end-users. We evaluate our CEO model on a cluster of more than 40 nodes. Our experiment results show that our CEO model is computationally flexible, scalable and practical.