![[2506.16404] Generating Directed Graphs with Dual Attention and Asymmetric Encoding](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2506.16404] Generating Directed Graphs with Dual Attention and Asymmetric Encoding
Summary
This article presents a novel generative model for directed graphs called Directo, which utilizes dual attention and asymmetric encoding to enhance graph generation capabilities, addressing key challenges in the field.
Why It Matters
Directed graphs are crucial for modeling complex systems in various domains. This research advances the understanding and generation of directed graphs, providing a new framework that could improve applications in data simulation, augmentation, and discovery.
Key Takeaways
- Directo is the first generative model specifically designed for directed graphs.
- The model employs dual attention to capture both incoming and outgoing dependencies effectively.
- A benchmark suite is introduced to evaluate the model's performance against existing methods.
- Directo shows strong performance across diverse datasets, competing with specialized models.
- The research lays a foundation for future advancements in directed graph generation.
Computer Science > Machine Learning arXiv:2506.16404 (cs) [Submitted on 19 Jun 2025 (v1), last revised 19 Feb 2026 (this version, v3)] Title:Generating Directed Graphs with Dual Attention and Asymmetric Encoding Authors:Alba Carballo-Castro, Manuel Madeira, Yiming Qin, Dorina Thanou, Pascal Frossard View a PDF of the paper titled Generating Directed Graphs with Dual Attention and Asymmetric Encoding, by Alba Carballo-Castro and 4 other authors View PDF HTML (experimental) Abstract:Directed graphs naturally model systems with asymmetric, ordered relationships, essential to applications in biology, transportation, social networks, and visual understanding. Generating such graphs enables tasks such as simulation, data augmentation and novel instance discovery; however, directed graph generation remains underexplored. We identify two key factors limiting progress in this direction: first, modeling edge directionality introduces a substantially larger dependency space, making the underlying distribution harder to learn; second, the absence of standardized benchmarks hinders rigorous evaluation. Addressing the former requires more expressive models that are sensitive to directional topologies. We propose Directo, the first generative model for directed graphs built upon the discrete flow matching framework. Our approach combines: (i) principled positional encodings tailored to asymmetric pairwise relations, (ii) a dual-attention mechanism capturing both incoming and outgoing dep...
