Vision GNNs (ViGs) divide an image into multiple patches, treating these image patches as graph nodes. The image is represented by extracting explicit features from these patches as node features and constructing edge connections based on explicit dependencies. However, this explicit graph structure struggles to accurately capture deeper implicit dependencies. For example, at the node-level, implicit relationships include the intra-group consistency of local and global features belonging to the same semantic group and the inter-group distinction of features belonging to different semantic groups. At the graph-level, implicit relationships manifest in whether global consistency of edge connections can be established in the absence of direct edge connection supervision. These aspects are crucial for improving the accuracy of downstream tasks. Therefore, more effective learning of implicit dependencies in vision graph structures remains an area requiring further research. We designed the Discriminative Feature Reorganization (DFR) module to address implicit dependencies at the node-level. This module constructs a loss function using similarity measures between positive and negative sample feature pairs from adjacent layers of the neural network. By adjusting this loss function, the intra-group consistency and inter-group distinction of node-level local and global features can be enhanced. We also designed the Graph Structure Refinement (GSR) module. This module refines the consistency of graph-level implicit relationships of edge connections through interactive supervision of two graphs learned from adjacent layers of the neural network. Experimental results show that ViDR-GNN achieves significant performance improvements in image classification, object detection, and instance segmentation tasks.

Detecting leakages in a water distribution network (WDN) is a challenging task due to the complexity of data patterns caused by the pipeline leakages and the volatility of the daily demands. Usually, the data under normal operations are collected and different machine learning algorithms are developed to predict anomalies due to the leaks. However, these methods are overwhelmingly rely on the time domain modeling and ignore the information in the frequency domain, and lack a comprehensive modeling of the data patterns such as shapelet, trend, seasonality and point outliers. In this paper, we propose a joint time-frequency attention (JTFA) approach to detect the WDN leakages. In essence, the received signals are decomposed into trend and residual components to represent the incipient and abrupt leaks separately. Attention models are then applied on both time and frequency domain signals to learn the corresponding patterns. In particular, the spectrum is divided into different frequency bands to better attend the detailed information in the higher frequency bands. The desired signals are subsequently reconstructed and compared to the input signal to generate an anomaly score. Experiments from simulated water supply networks are organized and the results demonstrate that the proposed approach performs better than existing leak detection methods and time-frequency analysis methods.

Many works on graph neural networks (GNNs) focus on graph topologies and analyze graph-related operations to enhance performance on tasks such as node classification. In this paper, we propose to understand GNNs based on a feature-centric approach. Our main idea is to treat the features of nodes from each label class as a whole, from which we can identify the centroid. The convex hull of these centroids forms a simplex called the feature centroid simplex, where a simplex is a high-dimensional generalization of a triangle. We borrow ideas from coarse geometry to analyze the geometric properties of the feature centroid simplex by comparing them with basic geometric models, such as regular simplexes and degenerate simplexes. Such a simplex provides a simple platform to understand graph-based feature aggregation, including phenomena such as heterophily, oversmoothing, and feature re-shuffling. Based on the theory, we also identify simple and useful tricks for the node classification task.

As the number of layers in Graph Neural Networks (GNNs) increases, over-smoothing becomes more severe, causing intra-class feature distances to shrink, while heterogeneous representations tend to converge. Most existing methods attempt to address this issue by employing heuristic shortcut mechanisms or optimizing objectives to constrain inter-class feature differences. However, these approaches fail to establish a theoretical connection between message passing and the variation in inter-class feature differences, making it challenging to design methods that target the key influencing factors. To address this gap, this paper first introduces the concept of the separation coefficient, which quantifies the contraction of feature distances between classes during multi-layer message passing. Based on this theory, we propose a low-complexity, pluggable, pseudo-label-based adaptive graph structure adjustment method. This approach effectively enhances the separation coefficient of inter-class features while maintaining intra-class compactness, thereby alleviating the convergence of heterogeneous representations caused by multi-layer aggregation. Experimental results demonstrate that the proposed method significantly improves the discriminability of node representations and enhances node classification performance across various datasets and foundational models.

Array-geometry-agnostic speech separation (AGA-SS) aims to develop an effective separation method regardless of the microphone array geometry. Conventional methods rely on permutation-free operations, such as summation or attention mechanisms, to capture spatial information. However, these approaches often incur high computational costs or disrupt the effective use of spatial information during intra- and inter-channel interactions, leading to suboptimal performance. To address these issues, we propose UniArray, a novel approach that abandons the conventional interleaving manner. UniArray consists of three key components: a virtual microphone estimation (VME) module, a feature extraction and fusion module, and a hierarchical dual-path separator. The VME ensures robust performance across arrays with varying channel numbers. The feature extraction and fusion module leverages a spectral feature extraction module and a spatial dictionary learning (SDL) module to extract and fuse frequency-bin-level features, allowing the separator to focus on using the fused features. The hierarchical dual-path separator models feature dependencies along the time and frequency axes while maintaining computational efficiency. Experimental results show that UniArray outperforms state-of-the-art methods in SI-SDRi, WB-PESQ, NB-PESQ, and STOI across both seen and unseen array geometries.

Source-Free Unsupervised Domain Adaptation (SFUDA) enables model adaptation to unlabeled target domains without accessing source data. However, when the source domain contains classes absent in the target domain, existing methods suffer from negative transfer: knowledge of irrelevant source-only classes interferes with target class recognition, significantly degrading classification accuracy. We propose Machine Unlearning-based SFUDA (MUSFUDA), which addresses this problem by selectively unlearning source-only class knowledge from the pre-trained model rather than adding compensatory mechanisms. This machine unlearning approach allows the model to focus on shared classes, fundamentally eliminating negative transfer. Remote sensing images with large intra-class variations and high inter-class similarity cause over-unlearning of target classes when forgetting source-only classes, thus we design the Model-Disruption Based Dual-Teacher Unlearning Strategy (MDUS), which uses dual teachers to manage target class preservation and source-only class erasure through knowledge distillation. MDUS is lightweight and easily integrated into existing SFUDA frameworks. Experiments on remote sensing datasets demonstrate that combining MDUS with representative baselines consistently reduces negative transfer and improves classification performance, maintaining high efficiency, validating the effectiveness and generalizability of our approach.

Due to inappropriate sample selection and limited training data, a distribution shift often exists between the training and test sets. This shift can adversely affect the test performance of graph neural networks (GNNs). Existing approaches mitigate this issue by either enhancing the robustness of GNNs to distribution shift or reducing the shift itself. However, both approaches necessitate retraining the model, which becomes unfeasible when the model structure and parameters are inaccessible. To address this challenge, we propose FR-GNN, a general framework for GNNs to conduct feature reconstruction. FR-GNN constructs a mapping relationship between the output and input of a well-trained GNN to obtain class representative embeddings and then uses these embeddings to reconstruct the features of labeled nodes. These reconstructed features are then incorporated into the message passing mechanism of GNNs to influence the predictions of unlabeled nodes at test time. Notably, the reconstructed node features can be directly utilized for testing the well-trained model, effectively reducing the distribution shift and leading to improved test performance. This remarkable achievement is attained without any modifications to the model structure or parameters. We provide theoretical guarantees for the effectiveness of our framework. Furthermore, we conduct comprehensive experiments on various public data sets. The experimental results demonstrate the superior performance of FR-GNN in comparison to multiple categories of baseline methods.

Graph adversarial attacks can be classified as either white-box or black-box attacks. White-box attackers typically exhibit better performance because they can exploit the known structure of victim models. However, in practical settings, most attackers generate perturbations under black-box conditions, where the victim model is unknown. A fundamental question is how to leverage a white-box attacker to attack a black-box model. Some current black-box attack approaches employ white-box techniques to attack a surrogate model, resulting in satisfactory outcomes. Nonetheless, such white-box attackers must be meticulously designed and lack theoretical assurances for attack effectiveness. In this paper, we propose a novel framework that utilizes simple white-box techniques to conduct black-box attacks and provides the lower bound for attack performance. Specifically, we first employ a more comprehensive GCN technique named BiasGCN to approximate the victim model, and subsequently, use a simple white-box approach to attack the approximate model. We provide a generalization guarantee for our BiasGCN and employ it to obtain the lower bound on attack performance. Our method is evaluated on various datasets, and the experimental results indicate that our approach surpasses recently proposed baselines.

Many graph neural networks (GNNs) are inapplicable when the graph structure representing the node relations is unavailable. Recent studies have shown that this problem can be effectively solved by jointly learning the graph structure and the parameters of GNNs. However, most of these methods learn graphs by using either a Euclidean or hyperbolic metric, which means that the space curvature is assumed to be either constant zero or constant negative. Graph embedding spaces usually have nonconstant curvatures, and thus, such an assumption may produce some obfuscatory nodes, which are improperly embedded and close to multiple categories. In this article, we propose a joint-space graph learning (JSGL) method for GNNs. JSGL learns a graph based on Euclidean embeddings and identifies Euclidean obfuscatory nodes. Then, the graph topology near the identified obfuscatory nodes is refined in hyperbolic space. We also present a theoretical justification of our method for identifying obfuscatory nodes and conduct a series of experiments to test the performance of JSGL. The results show that JSGL outperforms many baseline methods. To obtain more insights, we analyze potential reasons for this superior performance.

Graph neural networks (GNNs) have achieved great success in many fields due to their powerful capabilities of processing graph-structured data. However, most GNNs can only be applied to scenarios where graphs are known, but real-world data are often noisy or even do not have available graph structures. Recently, graph learning has attracted increasing attention in dealing with these problems. In this article, we develop a novel approach to improving the robustness of the GNNs, called composite GNN. Different from existing methods, our method uses composite graphs (C-graphs) to characterize both sample and feature relations. The C-graph is a unified graph that unifies these two kinds of relations, where edges between samples represent sample similarities, and each sample has a tree-based feature graph to model feature importance and combination preference. By jointly learning multiaspect C-graphs and neural network parameters, our method improves the performance of semisupervised node classification and ensures robustness. We conduct a series of experiments to evaluate the performance of our method and the variants of our method that only learn sample relations or feature relations. Extensive experimental results on nine benchmark datasets demonstrate that our proposed method achieves the best performance on almost all the datasets and is robust to feature noises.

In node classification using graph neural networks (GNNs), a typical model generates logits for different class labels at each node. A softmax layer often outputs a label prediction based on the largest logit. We demonstrate that it is possible to infer hidden graph structural information from the dataset using these logits. We introduce the key notion of label non-uniformity, which is derived from the Wasserstein distance between the softmax distribution of the logits and the uniform distribution. We demonstrate that nodes with small label non-uniformity are harder to classify correctly. We theoretically analyze how the label non-uniformity varies across the graph, which provides insights into boosting the model performance: increasing training samples with high non-uniformity or dropping edges to reduce the maximal cut size of the node set of small non-uniformity. These mechanisms can be easily added to a base GNN model. Experimental results demonstrate that our approach improves the performance of many benchmark base models.

The problem of estimating event truths from conflicting agent opinions in a social network is investigated. An autoencoder learns the complex relationships between event truths, agent reliabilities and agent observations. A Bayesian network model is proposed to guide the learning process by modeling the relationship of the autoencoder's outputs with different variables. At the same time, it also models the social relationships between agents in the network. The proposed approach is unsupervised and is applicable when ground truth labels of events are unavailable. A variational inference method is used to jointly estimate the hidden variables in the Bayesian network and the parameters in the autoencoder. Experiments on three real datasets demonstrate that our proposed approach is competitive with, and in most cases better than, several state-of-the-art benchmark methods.

The problem of multiple acoustic source localization using observations from a microphone array network is investigated in this article. Multiple source signals are assumed to be window-disjoint-orthogonal (WDO) on the time-frequency (TF) domain and time delay of arrival (TDOA) measurements are extracted at each TF bin. A Bayesian network model is then proposed to jointly assign the measurements to different sources and estimate the acoustic source locations. Considering that the WDO assumption is usually violated under reverberant and noisy environments, we construct a relational network by coding the distance information between the distributed microphone arrays such that adjacent arrays have higher probabilities of observing the same acoustic source, which is able to mitigate the miss detection issues in adverse environments. A Laplace approximate variational inference method is introduced to estimate the hidden variables in the proposed Bayesian network model. Both simulations and real data experiments are performed. The results show that our proposed method is able to achieve better source localization accuracy than existing methods.

In view of the huge success of convolution neural networks (CNN) for image classification and object recognition, there have been attempts to generalize the method to general graph-structured data. One major direction is based on spectral graph theory. In this paper, we study the problem from a different perspective, by introducing parallel flow decomposition of graphs. The essential idea is to decompose a graph into families of non-intersecting one dimensional (1D) paths, after which, we may apply a 1D CNN along each family of paths. We demonstrate that the our method, which we call GFCN (graph flow convolutional network), is able to transfer CNN architectures to general graphs. We demonstrate effectiveness of the method with synthetic and real applications.

We consider the problem of blind calibration of a sensor network, where the sensor gains and offsets are estimated from noisy observations of unknown signals. This is in general a nonidentifiable problem, unless restrictive assumptions on the signal subspace or sensor observations are imposed. We show that if each signal observed by the sensors follows a known dynamic model with additive noise, then the sensor gains and offsets are identifiable. We propose a dynamic Bayesian nonparametric model to infer the sensors' gains and offsets. Our model allows different sensor clusters to observe different unknown signals, without knowing the sensor clusters a priori. We develop an offline algorithm using block Gibbs sampling and a linearized forward filtering backward sampling method that estimates the sensor clusters, gains, and offsets jointly. Furthermore, for practical implementation, we also propose an online inference algorithm based on particle filtering and local Markov chain Monte Carlo. Simulations using a synthetic dataset, and experiments on two real datasets suggest that our proposed methods perform better than several other blind calibration methods, including a sparse Bayesian learning approach, and methods that first cluster the sensor observations and then estimate the gains and offsets.

In the sensor network blind calibration problem, the gains and offsets of sensors are estimated from noisy observations of unknown underlying signals. This is in general a non-identifiable problem, unless restrictive assumptions on the signal subspace or sensor observations are imposed. To overcome these assumptions, we propose a dynamic Bayesian nonparametric model. We show that if the unknown underlying signals follow the first-order auto-regressive process, then the sensor gains and offsets are identifiable. Furthermore, our model allows sensors to form clusters, where each cluster observes the same underlying signal. The clusters are however not known a priori, and are learned through the sensor data. We present a block Gibbs sampling inference method based on the forward filtering backward sampling algorithm. Simulation results suggest that our approach can estimate the sensor gains and offsets with good accuracy, and performs better than methods that first perform clustering and then blind calibration.