OpenNovelty.org

Graph Mixing Additive Networks

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 5.5Download Report PDF
Graph Deep LearningGraph Neural NetworksInterpretability
Abstract:

Real-world temporal data often consists of multiple signal types recorded at irregular, asynchronous intervals. For instance, in the medical domain, different types of blood tests can be measured at different times and frequencies, resulting in fragmented and unevenly scattered temporal data. Similar issues of irregular sampling occur in other domains, such as the monitoring of large systems using event log files. Effectively learning from such data requires handling sets of temporally sparse and heterogeneous signals. In this work, we propose Graph Mixing Additive Networks (GMAN), a novel and interpretable-by-design framework for learning directly from sets of graphs that represent such signals. GMAN provides diverse interpretability capabilities, including node-level, graph-level, and subset-level importance, and enables practitioners to trade finer-grained interpretability for greater expressivity when domain priors are available. GMAN achieves state-of-the-art performance in real-world high-stakes tasks, including predicting Crohn’s disease onset and hospital length of stay from routine blood test measurements and detecting fake news. Furthermore, we demonstrate how GMAN’s interpretability properties assist in revealing disease development phase transitions and provide crucial insights in the healthcare domain.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper proposes Graph Mixing Additive Networks (GMAN), a framework for learning from sets of graphs representing irregular temporal signals, with applications to medical time series and fake news detection. Within the taxonomy, it resides in the 'Healthcare and Medical Time Series Analysis' leaf under 'Domain-Specific Applications of Graph-Based Temporal Learning'. This leaf contains six papers total, indicating a moderately populated research direction. The sibling works include Graph Medical Analysis, Dynamic Graph Medical, and others addressing patient monitoring and clinical prediction from irregular health records, suggesting GMAN enters a space with established prior art.

The taxonomy reveals neighboring leaves focused on general-purpose architectures for irregular time series (e.g., message passing methods, neural ODEs) and broader temporal graph learning (e.g., continuous-time networks, self-supervised approaches). GMAN's positioning in a domain-specific leaf suggests it tailors general temporal graph techniques to healthcare constraints like interpretability and sparse observations. The taxonomy's scope note emphasizes medical diagnosis and patient monitoring, while excluding general temporal graph methods, clarifying that GMAN's healthcare focus differentiates it from purely methodological contributions in sibling branches like 'Graph Neural Network Architectures for Irregular Time Series'.

Among the three contributions analyzed, the literature search examined 23 candidates total, with no clearly refutable pairs identified. The GMAN framework itself was compared against 3 candidates, domain-prior integration against 10, and multi-level interpretability against 10, all yielding zero refutable overlaps. This suggests that within the limited search scope—top-K semantic matches plus citation expansion—no prior work directly anticipates GMAN's specific combination of graph mixing, additive structure, and interpretability mechanisms. However, the absence of refutations reflects the search scale (23 papers, not hundreds), leaving open the possibility of relevant work outside the examined set.

Given the moderately crowded healthcare leaf and the limited search scope, GMAN appears to offer a distinct approach within its niche, though the analysis does not cover exhaustive prior art. The taxonomy context indicates that while healthcare applications of temporal graph learning are active, GMAN's interpretability-by-design and domain-prior integration may represent incremental refinements rather than foundational shifts, pending broader literature review.

Taxonomy

Core-task Taxonomy Papers
50
3
Claimed Contributions
23
Contribution Candidate Papers Compared
0
Refutable Paper

Research Landscape Overview

Core task: Learning from sets of irregular temporal signals represented as graphs. This field addresses scenarios where observations arrive at non-uniform intervals and exhibit relational structure, requiring methods that jointly model temporal dynamics and graph topology. The taxonomy reveals six major branches: Graph Neural Network Architectures for Irregular Time Series focuses on neural designs that handle asynchronous observations and missing data, often employing continuous-time frameworks like neural ODEs (e.g., Temporal Graph ODEs[2], Mixed Graph NCDEs[6]). Temporal Graph Learning for Dynamic Networks emphasizes evolving connectivity patterns, with works such as Temporal Graph Networks[8] and Inductive Temporal Graphs[5] tackling link prediction and node classification on time-varying graphs. Temporal Knowledge Graph Reasoning and Forecasting targets symbolic relational data, where methods like Temporal KG Embedding[7] and Evolutional KG Reasoning[10] predict future facts. Spatio-Temporal Graph Forecasting and Imputation addresses structured prediction tasks in domains like traffic and climate, exemplified by Graph Time Forecasting[14] and Asynchronous Traffic Forecasting[24]. Domain-Specific Applications of Graph-Based Temporal Learning tailors these techniques to healthcare, environmental monitoring, and other specialized settings, while Foundational Graph Signal Processing and Theory provides the mathematical underpinnings, including classical signal processing perspectives like Graph Signal Overview[49]. Recent activity highlights contrasts between continuous-time modeling and discrete event-based approaches, as well as trade-offs between expressive power and computational efficiency. Within healthcare applications, Graph Mixing[0] sits alongside works like Graph Medical Analysis[4] and Dynamic Graph Medical[22], which similarly leverage patient interaction graphs and clinical time series. Compared to DynaGraph[40] and DynEdges-TGN[41], which emphasize dynamic edge modeling in general temporal graphs, Graph Mixing[0] focuses on mixing irregular temporal signals within a medical context, addressing challenges like sparse observations and heterogeneous patient trajectories. This positioning reflects a broader trend where domain-specific constraints—such as privacy, interpretability, and irregular sampling in healthcare—drive specialized architectures beyond generic temporal graph methods. Open questions remain around scalability to large patient cohorts, integration of multimodal data, and principled handling of missing or censored observations across diverse clinical settings.

Claimed Contributions

Graph Mixing Additive Networks (GMAN) framework

The authors propose GMAN, a new interpretable-by-design framework that operates on sets of graphs representing heterogeneous temporal signals with irregular sampling. Unlike existing methods that require alignment to fixed time grids or imputation, GMAN learns directly from the raw sparse data structure.

3 retrieved papers
Domain-prior integration via signal and feature grouping

The framework enables users to group related signals or features into subsets based on domain knowledge. This mechanism trades granular interpretability for increased model expressivity, which is particularly valuable in domains like medicine where such priors are common.

10 retrieved papers
Multi-level interpretability capabilities

GMAN offers interpretability at multiple granularities by design, providing importance scores at the node, graph, and subset levels. These scores directly reflect contributions to predictions and enable practitioners to understand how individual measurements, entire signals, or signal groups influence outcomes.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

Graph Mixing Additive Networks (GMAN) framework

The authors propose GMAN, a new interpretable-by-design framework that operates on sets of graphs representing heterogeneous temporal signals with irregular sampling. Unlike existing methods that require alignment to fixed time grids or imputation, GMAN learns directly from the raw sparse data structure.

Contribution

Domain-prior integration via signal and feature grouping

The framework enables users to group related signals or features into subsets based on domain knowledge. This mechanism trades granular interpretability for increased model expressivity, which is particularly valuable in domains like medicine where such priors are common.

Contribution

Multi-level interpretability capabilities

GMAN offers interpretability at multiple granularities by design, providing importance scores at the node, graph, and subset levels. These scores directly reflect contributions to predictions and enable practitioners to understand how individual measurements, entire signals, or signal groups influence outcomes.

Table of Contents