Research map / Graph learning / Graph self-supervised learning
NetDiff: Graph Diffusion with Improved Global Capabilities to Generate and Update Mobile Network Topologies
Félix Marcoccia, Victor Fagoo, Gilles Monzat de Saint Julien, Cédric Adjih, Thomas Watteyne, Paul Mühlethaler
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add domain-related features to the diffusion graph transformer (ablation) | DDPM-GT · 16-node graphs | saturation (%): 17.6 -> 13.6 | Improved |
| add softmax CAM tokens (ablation) | DDPM-GT w/ features · 16-node graphs | saturation (%): 13.6 -> 12.9 | Improved |
| use unsoftmaxed absolute attention (ACAM) instead of softmax CAM (ablation) | NetDiff · 16-node graphs | saturation (%): 12.9 -> 12.1 | Improved |
| use ACAM instead of CAM tokens (ablation) | NetDiff · 16-node graphs | connected (%) and parity (%): 98.68 -> 98.68 connected; 98.63 -> 98.64 parity | No clear effect |
| add CAM tokens (ablation) | DDPM-GT · 16-node graphs | node saturation (%): 11.62 -> 7.98 | Improved |
| add ACAM tokens (ablation) | DDPM-GT · 16-node graphs | node saturation (%): 11.62 -> 7.05 | Improved |
| use the added sector, cosine and parity losses instead of standard BCE (ablation) | NetDiff (ACAM) · 16-node graphs | parity (%): 62.23 -> 98.64 | Improved |
| use the added losses instead of standard BCE (ablation) | NetDiff (ACAM) · 16-node graphs | node saturation (%): 15.90 -> 7.05 | Improved |
| increase ACAM tokens from 1 to 16 (ablation) | NetDiff · 16-node graphs | saturation metrics: reduces saturation by approximately 0.8% | Improved |
| use NetDiff topologies instead of DDPM-GT | NetDiff · 32-node networks | throughput (Mbps): 258.90 -> 322.40 | Improved |
| use partial diffusion for reconfiguration instead of full generation | NetDiff · 16-node graphs, mobility | similarity to previous topology: about 21 % more similar | Improved |
| use noise-less minor reconfiguration with 7 to 10 steps | NetDiff · 16-node graphs, minor movement | similarity to previous topology: 41 % more similar | Improved |
Closest papers by meaning
- When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors (NeurIPS 2026 · similarity 0.84)
- Is Noise Conditioning Necessary? A Unified Theory of Unconditional Graph Diffusion Models (NeurIPS 2025 · similarity 0.84)
- Pard: Permutation-Invariant Autoregressive Diffusion for Graph Generation (NeurIPS 2024 · similarity 0.84)
- Generating Graphs via Spectral Diffusion (ICLR 2025 · similarity 0.82)
- Simple and Critical Iterative Denoising: A Recasting of Discrete Diffusion in Graph Generation (ICML 2025 · similarity 0.82)
- Communication-Efficient Diffusion Denoising Parallelization via Reuse-then-Predict Mechanism (NeurIPS 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
