Efficient Fair Graph Representation Learning Using a Multi-level Framework

Cross-scale contrastive triplet networks for graph representation learning

Graph representation for structural topology optimization using genetic algorithms

Cross-view graph contrastive learning with hypergraph

Recently, novel technologies have enabled spatially resolved multiplexed protein profiling of tumors; however, there is an unmet need for unbiased detection of spatial structures within the tissue. By encoding multiplexed images as a graph, we produce an efficient and well-studied graph representation that allows for the application of sophisticated machine learning methods, particularly graph neural networks (GNNs) and graph autoencoders (GAE). While GNNs have recently become popular tools in the analysis of spatial data, current GNN methods are limited by production of cell-level representations rather than graph-level representations. To understand patterns of local tissue structure and protein expression, we implemented a generalizable GAE framework that summarizes marker intensity and morphological values with the local topology of neighboring cells. In this way, each local cellular graph is represented as a single point in the lower dimensional space. We show that our framework generates low-dimensional features that reflect both cellular attributes and tissue structure. We demonstrate that clustering local structure in 318 fields of view from 22 tumor lesions across 7 patients with in-transit melanoma captured using multiplexed immunofluorescence (mpIF) can produce spatial motifs predictive of response to intralesional interleukin-2 (IL-2) immunotherapy. Additionally, we applied our framework to mpIF data of 146 high grade serous ovarian cancer tumors across 50 patients. We identify clusters with higher CD8+ T cell infiltration that associate with homologous recombination deficiency. Our method can be readily applied to various spatially resolved multiplexed imaging platforms, and the resulting embeddings provide a compressed representation of subgraphs that can be used in further classification or regression analysis of patient- and/or image-level histological, clinical, or genomic labels. Citation Format: Nicholas Ceglia, Samuel S. Freeman, Maryam Pourmaleki, Andrew McPherson, Sohrab P. Shah. Graph representation learning of tumor topology from spatial imaging data. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5353.

In the field of computer vision, Vision Graph Neural Networks (ViG) have demonstrated significant potential in image understanding. By treating the divided image patches as nodes and constructing connection relationships based on neighbor attributes, ViG can efficiently model global dependencies within images with the help of graph attributes. However, most existing ViG methods have the problem of high computational complexity in graph construction and may not be able to effectively and fully explore the graph structure information. Besides, the heavy reliance on manual annotation labels limits the application potential of ViG in practical scenarios. To this end, in this paper, we propose a novel self-supervised vision graph contrastive learning method (S2ViG) based on image mixing strategy for efficient vision graph representation learning. It aims to use self-supervised method to alleviate the dependence on manual annotation and enhance the understanding of the global structure of the graph using two different vision graph construction methods. Specifically, we first employ image mixing strategy to uncover latent semantic relationships among multiple images. Then, we construct dynamic graph structures for image patches from local and global perspectives to obtain augmented contrastive samples. Finally, the multilevel contrastive loss is constructed to optimize the network. Experimental results show that our method achieves excellent performance on multiple datasets such as ImageNet-1K and CIFAR.

The prevailing graph neural network models have achieved significant progress in graph representation learning. However, in this paper, we uncover an ever-overlooked phenomenon: the pre-trained graph representation learning model tested with full graphs underperforms the model tested with well-pruned graphs. This observation reveals that there exist confounders in graphs, which may interfere with the model learning semantic information, and current graph representation learning methods have not eliminated their influence. To tackle this issue, we propose Robust Causal Graph Representation Learning (RCGRL) to learn robust graph representations against confounding effects. RCGRL introduces an active approach to generate instrumental variables under unconditional moment restrictions, which empowers the graph representation learning model to eliminate confounders, thereby capturing discriminative information that is causally related to downstream predictions. We offer theorems and proofs to guarantee the theoretical effectiveness of the proposed approach. Empirically, we conduct extensive experiments on a synthetic dataset and multiple benchmark datasets. Experimental results demonstrate the effectiveness and generalization ability of RCGRL. Our codes are available at https://github.com/hang53/RCGRL.

Graph representation learning (GRL) aims to map a graph into a low-dimensional vector space while preserving graph topology and node properties. This study proposes a novel GRL model, Multi-Level GRL (simply, ML-GRL), that recursively partitions input graphs by selecting the most appropriate community detection algorithm at each graph or partitioned subgraph. To preserve the relationship between subgraphs, ML-GRL incorporates global graphs that effectively maintain the overall topology. ML-GRL employs a prediction model, which is pre-trained using graph-based features and covers a wide range of graph distributions, to estimate GRL accuracy of each community detection algorithm without partitioning graphs or subgraphs and evaluating them. ML-GRL improves learning accuracy by selecting the most effective community detection algorithm while enhancing learning efficiency from parallel processing of partitioned subgraphs. Through extensive experiments with two different tasks, we demonstrate ML-GRL's superiority over the six representative GRL models in terms of both learning accuracy and efficiency. Specifically, ML-GRL not only improves the accuracy of existing GRL models by 3.68 ~ 47.59% for link prediction and 1.75 ~ 40.90% for node classification but also significantly reduces their running time by 9.63 ~ 62.71% and 7.14 ~ 82.14%, respectively. Our source code is available at https://github.com/pnpy6elp/Multi_Level_GRL.

The aim of webly supervised fine-grained image recognition (FGIR) is to distinguish sub-ordinate categories based on data retrieved from the Internet, which can significantly mitigate the dependence of deep learning on manually annotated labels. Most current fine-grained image recognition algorithms use a large-scale data-driven deep learning paradigm, which relies heavily on manually annotated labels. However, there is a large amount of weakly labeled free data on the Internet. To utilize fine-grained web data effectively, this paper proposes a Graph Representation and Metric Learning (GRML) framework to learn discriminative and effective holistic–local features by graph representation for web fine-grained images and to handle noisy labels simultaneously, thus effectively using webly supervised data for training. Specifically, we first design an attention-focused module to locate the most discriminative region with different spatial aspects and sizes. Next, a structured instance graph is constructed to correlate holistic and local features to model the holistic–local information interaction, while a graph prototype that contains both holistic and local information for each category is introduced to learn category-level graph representation to assist in processing the noisy labels. Finally, a graph matching module is further employed to explore the holistic–local information interaction through intra-graph node information propagation as well as to evaluate the similarity score between each instance graph and its corresponding category-level graph prototype through inter-graph node information propagation. Extensive experiments were conducted on three webly supervised FGIR benchmark datasets, Web-Bird, Web-Aircraft and Web-Car, with classification accuracy of 76.62%, 85.79% and 82.99%, respectively. In comparison with Peer-learning, the classification accuracies of the three datasets separately improved 2.47%, 4.72% and 1.59%.

Cybersecurity has become an important part of our daily lives. As an important part, there are many researches on intrusion detection based on host system call in recent years. Compared to sentences, a sequence of system calls has unique characteristics. It contains implicit pattern relationships that are less sensitive to the order of occurrence and that have less impact on the classification results when the frequency of system calls varies slightly. There are also various properties such as resource consumption, execution time, predefined rules, and empirical weights of system calls. Commonly used word embedding methods, such as Bow, TI-IDF, N-Gram, and Word2Vec, do not fully exploit such relationships in sequences as well as conveniently support attribute expansion. To solve these problems, we introduce Graph Representation based Intrusion Detection (GRID), an intrusion detection framework based on graph representation learning. It captures the potential relationships between system calls to learn better features, and it is applicable to a wide range of back-end classifiers. GRID utilizes a new sequence embedding method Graph Random State Embedding (GRSE) that uses graph structures to model a finite number of sequence items and represent the structural association relationships between them. A more efficient representation of sequence embeddings is generated by random walks, word embeddings, and graph pooling. Moreover, it can be easily extended to sequences with attributes. Our experimental results on the AFDA-LD dataset show that GRID has an average improvement of 2% using the GRSE embedding method comparing to others.

Graph representation learning has become a ubiquitous component in many scenarios, ranging from social network analysis to energy forecasting in smart grids. In several applications, ensuring the fairness of the node (or graph) representations with respect to some protected attributes is crucial for their correct deployment. Yet, fairness in graph deep learning remains underexplored, with few solutions available. In particular, the tendency of similar nodes to cluster on several real-world graphs (i.e., homophily) can dramatically worsen the fairness of these procedures. In this article, we propose a novel biased edge dropout algorithm (FairDrop) to counter-act homophily and improve fairness in graph representation learning. FairDrop can be plugged in easily on many existing algorithms, is efficient, adaptable, and can be combined with other fairness-inducing solutions. After describing the general algorithm, we demonstrate its application on two benchmark tasks, specifically, as a random walk model for producing node embeddings, and to a graph convolutional network for link prediction. We prove that the proposed algorithm can successfully improve the fairness of all models with only a small or negligible drop in accuracy, and compares favourably with existing state-of-the-art solutions. In an ablation study, we demonstrate that our algorithm can flexibly interpolate between biasing towards fairness and an unbiased edge dropout. Furthermore, to better evaluate the gains, we propose a new dyadic group definition to measure the bias of a link prediction task when paired with group-based fairness metrics. In particular, we extend the metric used to measure the bias in node embeddings by taking account of the graph structure.

Unsupervised multi-view graph representation learning with dual weight-net

A large number of real-world systems generate graphs that are structured data aligned with nodes and edges. Graphs are usually dynamic in many scenarios, where nodes or edges keep evolving over time. Recently, graph representation learning (GRL) has received great success in network analysis, which aims to produce informative and representative features or low-dimensional embeddings by exploring node attributes and network topology. Most state-of-the-art models for dynamic GRL are composed of a static representation learning model and a recurrent neural network (RNN). The former generates the representations of a graph or nodes from one static graph at a discrete time step, while the latter captures the temporal correlation between adjacent graphs. However, the two-stage design ignores the temporal dynamics between contiguous graphs during the learning processing of graph representations. To alleviate this problem, this article proposes a representation learning model for dynamic graphs, called DynGNN. Differently, it is a single-stage model that embeds an RNN into a graph neural network to produce better representations in a compact form. This takes the fusion of temporal and topology correlations into account from low-level to high-level feature learning, enabling the model to capture more fine-grained evolving patterns. From the experimental results on both synthetic and real-world networks, the proposed DynGNN yields significant improvements in multiple tasks compared to the state-of-the-art counterparts.

Semi-supervised learning has shown its potential in many real-world applications where only few labeled examples are available. However, when some fairness constraints need to be satisfied, semi-supervised classification models often struggle as they are required to cope with the lack of sufficient information for predicting the target variable while forgetting its relationships with any sensitive and potentially discriminatory attribute. To address this issue, we propose a fair semi-supervised representation learning architecture that leads to fair and accurate classification results even in very challenging scenarios with few labeled (but biased) instances. We show experimentally that our model can be easily adopted in very general settings, as the learned representations may be employed to train any supervised classifier. Moreover, when applied to several synthetic and real-world datasets, our method is competitive with state-of-the-art fair semi-supervised approaches.

With the development of artificial intelligence, the technology of graph representation has been widely used in graph structure data mining. The traditional graph representation mostly uses adjacency matrix. The representation of adjacency matrix is faced with two major challenges. One reason is extremely inefficient in the storage and computing of large graphs with the massive growth of data; the other is that the characterized data is difficult to retain the original structural information of graphs. Aiming at the above two points, an efficient graph representation method is proposed in this paper. The first-order similarity and second-order similarity of the graph are used as constraints to train the original graph structure data, and the nodes in the graph are mapped to a low-dimensional space, which greatly reduces the cost of storage space and computing while retaining the graph structure information. Experimental results show that the proposed method is significantly better than the Deepwalk and spectral clustering methods in different proportion of label data. Meanwhile, the proposed method can be applied to the semi-supervised learning task with a small number of labels, which provides convenience for the subsequent data analysis.

Graph similarity learning, which measures the similarities between a pair of graph-structured objects, lies at the core of various machine learning tasks such as graph classification, similarity search, etc. In this paper, we devise a novel graph neural network based framework to address this challenging problem, motivated by its great success in graph representation learning. As the vast majority of existing graph neural network models mainly concentrate on learning effective node or graph level representations of a single graph, little effort has been made to jointly reason over a pair of graph-structured inputs for graph similarity learning. To this end, we propose Hierarchical Hypergraph Matching Networks (H2sup>MN) to calculate the similarities between graph pairs with arbitrary structure. Specifically, our proposed H2MN learns graph representation from the perspective of hypergraph, and takes each hyperedge as a subgraph to perform subgraph matching, which could capture the rich substructure similarities across the graph. To enable hierarchical graph representation and fast similarity computation, we further propose a hyperedge pooling operator to transform each graph into a coarse graph of reduced size. Then, a multi-perspective cross-graph matching layer is employed on the coarsened graph pairs to extract the inter-graph similarity. Comprehensive experiments on five public datasets empirically demonstrate that our proposed model can outperform state-of-the-art baselines with different gains for graph-graph classification and regression tasks.