Structure and Organization

  • Chair: Prof. Vincenzo Carletti
  • Vice-chair: Prof. Benoit Gaüzère
  • Steering Committee: Prof. Mario Vento, Prof. Pasquale Foggia, Prof. Walter G. Kropatsch, Prof. Luc Brun, Prof. Xiaoyi Jiang

Contacts

  • Chair: chair@iapr-tc15.org
  • Steering Committee: steering@iapr-tc15.org

Aims and Scope

TC15 focuses on the theory, methodology, and applications of graph-based representations and representation learning in pattern recognition and related areas. Graphs are powerful tools for modeling data that exhibit structured, relational, or irregular properties. They are widely used to represent spatial relationships, topological configurations, semantic structures, and temporal dynamics.

In recent years, the scope of graph-based approaches has significantly expanded with the integration of machine learning techniques, particularly graph neural networks (GNNs) and geometric deep learning. These developments have enabled the automatic learning of expressive representations from graph-structured data, opening new avenues for analysis, classification, prediction, and synthesis.

TC15 aims to support and coordinate research activities in both foundational aspects and practical applications of graph-based methods. The technical scope includes:

Core Research Topics

  • Graph-based representations in pattern recognition
  • Graph neural networks and geometric deep learning
  • Graph matching, alignment, and edit distance
  • Graph-based clustering, classification, and retrieval
  • Graph embeddings and kernel methods
  • Spectral graph theory and Laplacian-based techniques
  • Graph similarity, distance measures, and topological descriptors
  • Learning on structured, heterogeneous, and attributed graphs
  • Dynamic and temporal graphs, including spatiotemporal modeling
  • Graph generation, synthesis, and probabilistic modeling
  • Explainability, interpretability, and uncertainty in graph-based learning
  • Topological data analysis and persistent homology on graphs

Application Domains

  • Computer vision and image analysis
  • Medical imaging and bioinformatics (e.g., molecular graphs, connectomics)
  • Remote sensing and geospatial analysis
  • Social network analysis and knowledge graphs
  • Robotics and autonomous systems (e.g., scene graphs, SLAM)
  • Human behavior and activity recognition
  • Document analysis and handwriting recognition
  • Cybersecurity and threat intelligence
  • Natural language processing with syntactic and semantic graphs

TC15 supports interdisciplinary collaboration and emphasizes both the development of new theoretical models and the translation of graph-based methods into robust, real-world applications. It encourages contributions that improve the scalability, interpretability, and reproducibility of graph-based systems.

Our Challenges

TC15 is committed to identifying and addressing key research challenges that can unify the efforts of the community and drive innovation in graph-based pattern recognition. These challenges highlight current limitations, open questions, and opportunities for cross-disciplinary collaboration.

1. Scalable Graph Matching and Analysis

Despite progress in graph matching, scalability remains a critical issue when dealing with large graphs. Many state-of-the-art algorithms struggle with graphs containing thousands of nodes and complex attributes.

Key questions:

  • How can we design matching algorithms with linear or sub-quadratic complexity?
  • Can we leverage graph sparsity, decomposition, or hierarchical structures?
  • How can graph coarsening, pruning, or attention mechanisms improve scalability?

2. Representation Learning on Graphs

Learning effective representations from graph-structured data is a central challenge, particularly in low-data, noisy, or heterogeneous settings.

Key directions:

  • Development of unsupervised and self-supervised learning frameworks
  • Transfer learning and domain adaptation across different graph domains
  • Learning representations that capture semantics, topology, and hierarchy

3. Enhancing Semantics in Low-Level Graphs

Graphs derived from images (e.g., Region Adjacency Graphs) often capture only low-level features. Enhancing their semantic content without compromising their geometric structure is crucial for tasks like recognition, parsing, or scene understanding.

Key directions:

  • Enriching image-based graphs with learned or symbolic attributes
  • Integrating multi-scale or multi-modal features
  • Combining pixel-level and object-level reasoning through hybrid graphs

4. Benchmarking, Evaluation, and Reproducibility

Robust benchmarking is essential for tracking progress and comparing methods. Yet, the field still lacks comprehensive, standardized benchmarks covering the diversity of graph problems.

Needs include:

  • Public datasets for graph matching, classification, segmentation, etc.
  • Agreed-upon metrics and evaluation protocols
  • Shared baselines and reproducible implementations

5. Spatiotemporal and High-Dimensional Graphs

Graph modeling is increasingly applied to time-varying or high-dimensional data — such as video sequences, dynamic social networks, or sensor streams.

Open challenges:

  • How can we model evolving graph structures over time?
  • What are the best representations for 3D, 4D (spatiotemporal), or even higher-dimensional graphs?
  • How can temporal dependencies and topological changes be learned jointly?

6. Explainability and Trust in Graph Learning

As graph-based models are adopted in critical applications, their transparency and robustness become essential.

Important topics:

  • Interpretable graph neural networks
  • Graph model auditing and debugging tools
  • Formal guarantees and uncertainty quantification

Future Activities

TC15 will continue to support and expand activities that promote excellence in research, foster collaboration, and increase the impact of graph-based methods in pattern recognition and beyond. Our future directions are organized around the following strategic goals:

1. Open Research and Reproducibility

We aim to promote transparency and rigor in the development and evaluation of graph-based techniques.

Key actions:

  • Support the creation and dissemination of publicly available datasets and code
  • Encourage the use of reproducibility checklists and open benchmarks in conferences
  • Facilitate collaborative platforms for sharing models, pipelines, and evaluations

2. Interdisciplinary Outreach and Collaboration

Graph-based models are increasingly relevant across domains such as biology, neuroscience, chemistry, social sciences, and cybersecurity. TC15 will encourage cross-disciplinary exchange to foster innovation.

Key actions:

  • Organize joint workshops and special sessions at major conferences
  • Establish collaborations with other IAPR TCs and external research communities
  • Promote application-focused research that bridges theory and practice

3. Education and Community Development

We recognize the importance of training and supporting the next generation of researchers in graph-based methods.

Key actions:

  • Develop tutorials and lecture series on foundational and emerging topics (e.g., GNNs, spectral methods, topological data analysis)
  • Support summer schools and mentorship programs
  • Promote diversity, equity, and inclusion in community initiatives