Most graph neural network models learn embeddings of nodes in static attributed graphs for predictive analysis. Recent attempts have been made to learn temporal proximity of the nodes. We find that real dynamic attributed graphs exhibit complex phenomenon of co-evolution between node attributes and graph structure. Learning node embeddings for forecasting change of node attributes and evolution of graph structure over time remains an open problem. In this work, we present a novel framework called CoEvoGNN for modeling dynamic attributed graph sequence. It preserves the impact of earlier graphs on the current graph by embedding generation through the sequence of attributed graphs. It has a temporal self-attention architecture to model long-range dependencies in the evolution. Moreover, CoEvoGNN optimizes model parameters jointly on two dynamic tasks, attribute inference and link prediction over time. So the model can capture the co-evolutionary patterns of attribute change and link formation. This framework can adapt to any graph neural algorithms so we implemented and investigated three methods based on it: CoEvoGCN, CoEvoGAT, and CoEvoSAGE. Experiments demonstrate the framework (and its methods) outperforms strong baseline methods on predicting an entire unseen graph snapshot of personal attributes and interpersonal links in dynamic social graphs and financial graphs.
Asymptotic distribution-free changepoint detection for data with repeated observations
Summary A nonparametric framework for changepoint detection, based on scan statistics utilizing graphs that represent similarities among observations, is gaining attention owing to its flexibility and good performance for high-dimensional and non-Euclidean data sequences. However, this graph-based framework faces challenges when there are repeated observations in the sequence, which is often the case for discrete data such as network data. In this article we extend the graph-based framework to solve this problem by averaging or taking the union of all possible optimal graphs resulting from repeated observations. We consider both the single-changepoint alternative and the changed-interval alternative, and derive analytical formulas to control the Type I error for the new methods, making them readily applicable to large datasets. The extended methods are illustrated on an application in detecting changes in a sequence of dynamic networks over time. All proposed methods are implemented in an $\texttt{R}$ package $\texttt{gSeg}$ available on CRAN.
- Publication Date:
- NSF-PAR ID:
- 10334651
- Journal Name:
- Biometrika
- ISSN:
- 0006-3444
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Few-shot node classification is tasked to provide accurate predictions for nodes from novel classes with only few representative labeled nodes. This problem has drawn tremendous attention for its projection to prevailing real-world applications, such as product categorization for newly added commodity categories on an E-commerce platform with scarce records or diagnoses for rare diseases on a patient similarity graph. To tackle such challenging label scarcity issues in the non-Euclidean graph domain, meta-learning has become a successful and predominant paradigm. More recently, inspired by the development of graph self-supervised learning, transferring pretrained node embeddings for few-shot node classification could be a promising alternative to meta-learning but remains unexposed. In this work, we empirically demonstrate the potential of an alternative framework, \textit{Transductive Linear Probing}, that transfers pretrained node embeddings, which are learned from graph contrastive learning methods. We further extend the setting of few-shot node classification from standard fully supervised to a more realistic self-supervised setting, where meta-learning methods cannot be easily deployed due to the shortage of supervision from training classes. Surprisingly, even without any ground-truth labels, transductive linear probing with self-supervised graph contrastive pretraining can outperform the state-of-the-art fully supervised meta-learning based methods under the same protocol. We hope thismore »
-
The key role of emotions in human life is undeniable. The question of whether there exists a brain pattern associated with a specific emotion is the theme of many affective neuroscience studies. In this work, we bring to bear graph signal processing (GSP) techniques to tackle the problem of automatic emotion recognition using brain signals. GSP is an extension of classical signal processing methods to complex networks where there exists an inherent relation graph. With the help of GSP, we propose a new framework for learning class-specific discriminative graphs. To that end, firstly we assume for each class of observations there exists a latent underlying graph representation. Secondly, we consider the observations are smooth on their corresponding class-specific sough graph while they are non-smooth on other classes’ graphs. The learned class-specific graph-based representations can act as sub-dictionaries and be utilized for the task of emotion classification. Applying the proposed method on an electroencephalogram (EEG) emotion recognition dataset indicates the superiority of our framework over other state-of-the-art methods.
-
Alkan, Can (Ed.)Abstract Motivation Pangenome variation graphs model the mutual alignment of collections of DNA sequences. A set of pairwise alignments implies a variation graph, but there are no scalable methods to generate such a graph from these alignments. Existing related approaches depend on a single reference, a specific ordering of genomes or a de Bruijn model based on a fixed k-mer length. A scalable, self-contained method to build pangenome graphs without such limitations would be a key step in pangenome construction and manipulation pipelines. Results We design the seqwish algorithm, which builds a variation graph from a set of sequences and alignments between them. We first transform the alignment set into an implicit interval tree. To build up the variation graph, we query this tree-based representation of the alignments to reduce transitive matches into single DNA segments in a sequence graph. By recording the mapping from input sequence to output graph, we can trace the original paths through this graph, yielding a pangenome variation graph. We present an implementation that operates in external memory, using disk-backed data structures and lock-free parallel methods to drive the core graph induction step. We demonstrate that our method scales to very large graph induction problemsmore »
-
Federated learning has emerged as an important paradigm for training machine learning models in different domains. For graph-level tasks such as graph classification, graphs can also be regarded as a special type of data samples, which can be collected and stored in separate local systems. Similar to other domains, multiple local systems, each holding a small set of graphs, may benefit from collaboratively training a powerful graph mining model, such as the popular graph neural networks (GNNs). To provide more motivation towards such endeavors, we analyze real-world graphs from different domains to confirm that they indeed share certain graph properties that are statistically significant compared with random graphs. However, we also find that different sets of graphs, even from the same domain or same dataset, are non-IID regarding both graph structures and node features. To handle this, we propose a graph clustered federated learning (GCFL) framework that dynamically finds clusters of local systems based on the gradients of GNNs, and theoretically justify that such clusters can reduce the structure and feature heterogeneity among graphs owned by the local systems. Moreover, we observe the gradients of GNNs to be rather fluctuating in GCFL which impedes high-quality clustering, and design a gradientmore »
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Publisher's Version of Record
- https://doi.org/10.1093/biomet/asab048
