A new arXiv paper examines end-to-end supervised graph prediction, where comparing predicted graphs to targets with arbitrary node orderings requires a permutation-invariant loss. The authors analyze three optimal transport-based graph matching relaxations and propose amortization to reduce the cost of these matching steps.
A new arXiv paper introduces GSLAD, an unsupervised method for detecting anomalies in multivariate time series. The approach focuses on changes in the structural relationships between variables, which often appear before individual readings deviate, and regularizes learned graph structures with prototypes. The authors argue this addresses a gap in forecasting- and reconstruction-based detection methods.
Researchers posted an arXiv paper describing a dataset and machine learning model for estimating water surface elevation across river networks where gauge coverage is sparse. The work frames river systems as a large spatiotemporal graph and targets applications such as flood forecasting and water resource management. The paper appeared as a new submission and has since been revised.
Researchers present a technique for exactly recovering communities in sparse directed stochastic block models by smoothing estimated outgoing connection-probability profiles before clustering vertices. The approach targets graphs where edges carry direction and connections are rare relative to the number of possible pairs. The work is a revised preprint posted on arXiv.
A newly published feature article lays out the mathematical underpinnings of covariance neural networks, a class of graph neural networks that treat covariance matrices as graph structures. The work connects classical dimensionality-reduction techniques such as PCA with modern graph-based learning, and highlights the broad range of domains where covariance data naturally occurs.
A revised arXiv preprint reviews how optimal transport is used to compare networks, offering both a dissimilarity measure and a plan describing how one graph maps onto another. The authors add new spectral bounds and discuss applications in machine learning. The paper is a cross-listing revision rather than a new release.
A newly posted arXiv paper introduces a method for adapting graph-based prediction models to new data with limited supervision or feedback. The approach, called reliability-aware prototype learning, aims to make agentic systems more data-efficient when reusing prior knowledge after deployment. It falls under the machine learning category and is a preprint, not yet peer-reviewed.
A revised arXiv paper presents a framework for inferring time-varying graphs from spatiotemporal measurements. By assuming a prior on how signals change over time, the method estimates changing graph structures even when only a small number of samples are available. The work targets settings where relationships between nodes shift dynamically rather than staying fixed.
A new arXiv paper proposes a defense for graph neural networks that strips out attacker-inserted edges from a graph's topology before inference, requiring no retraining of the model. The method relies on kernel-complexity signals to identify edges likely introduced through adversarial manipulation, and the authors support the design with theoretical guarantees. It addresses threats that target graph structure rather than node features.
A paper on arXiv investigates how well shallow graph embedding methods for community detection hold up when networks undergo perturbations, focusing on node deletions. It evaluates whether low-dimensional node representations preserve community structure as the underlying graph changes. The work offers insight into when these embedding techniques remain reliable in practice.
A newly posted arXiv paper proposes a multi-agent extension of agentic graph learning, a technique where LLM-driven agents sample parts of a graph as evidence before making a prediction. The approach uses structural signatures to coordinate how multiple agents divide and process graph reasoning tasks. The work targets improved performance on graph reasoning benchmarks compared with prior single-agent methods.