Research map / Graph learning / Graph self-supervised learning
FAP-CD: Fairness-Driven Age-Friendly Community Planning via Conditional Diffusion Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes FAP-CD, a conditional graph denoising diffusion model with a fairness-driven demand pre-training module that generates age-friendly community facility layouts in Beijing.
Problem
Age-friendly community facilities in cities are inadequate and unevenly distributed, and existing automated planning methods neglect elderly needs and spatial equity.
Approach
Frames age-friendly facility planning as conditional discrete graph generation: a graph denoising diffusion model conditioned on a min-max fairness demand embedding (attention-based) and a 15-minute walkable adjacency graph, with a hybrid graph transformer plus message-passing denoiser and node/edge augmentation.
Key results
- FAP-CD achieves Average score 0.588 versus 0.416 for the best baseline (DDPM), with Life Service 0.888 and Elderly Care 0.923.
- Accessibility improved by over 29% and Gini reduced by over 28% relative to the best baseline, per the text.
- Ablation: residual hybrid layer removal affects accessibility by up to 105.1%; augmentation module yields up to 13.1% Gini improvement.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use FAP-CD instead of DDPM | Age-friendly community planning · Beijing | Average: 0.416 -> 0.588 | Helped |
| Add the residual hybrid layer (ablation) | FAP-CD denoising network · Beijing | Accessibility: up to a 105.1% improvement | Helped |
| Add the graph transformer block (ablation) | FAP-CD denoising network · Beijing | AFC generation performance | Helped |
| Add the message passing block (ablation) | FAP-CD denoising network · Beijing | AFC generation performance | Helped |
| Add the node and edge attribute enhancement module (ablation) | FAP-CD diffusion model · Beijing | Gini: up to a 13.1% improvement | Helped |
| Use three residual hybrid layers rather than an overly deep network (ablation) | FAP-CD conditional graph denoising network · Beijing | Performance | Mixed |
| Increase hidden dimension to 256 rather than use 128 (ablation) | FAP-CD · Beijing | Performance | No clear effect |
Surprising findings
- Traditional DRF achieves the highest diversity among some baselines, attributed to stochastic and demand-driven allocation in ACA and GA.
- EDGE underperforms DDPM and the proposed model, suggesting MLP/GRU denoisers are weaker for this graph generation task.
Limitations
- Evaluated only on Beijing data at 2 km grid resolution.
- Accessibility gains vary widely (std 0.103), and fairness is measured with proxy metrics rather than real outcomes.
Experimental setup
Datasets: OpenStreetMap road networks Baidu Maps POIs Beijing senior meal program data Beijing senior care center data Lianjia residential area data 2020 Seventh National Population Census
Benchmarks: Beijing age-friendly facility layout dataset
Metrics: Life Service efficiency Elderly Care efficiency Diversity Accessibility Gini Average
Baselines: Walking-based ACA GA DRF VGAE GraphRNN CondGEN DDPM EDGE
Builds on: DDPM score-based SDE diffusion EDGE DPM-Solver graph transformer
Closest papers by meaning
- Monte Carlo Tree Diffusion for System 2 Planning (ICML 2025 · similarity 0.83)
- Resisting Stochastic Risks in Diffusion Planners with the Trajectory Aggregation Tree (ICML 2024 · similarity 0.81)
- When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors (NeurIPS 2026 · similarity 0.81)
- DiffuserLite: Towards Real-time Diffusion Planning (NeurIPS 2024 · similarity 0.80)
- Text2City: One-Stage Text-Driven Urban Layout Regeneration (AAAI 2024 · similarity 0.80)
- DiRecT: Safe Diffusion-Based Planning via Receding-Horizon Denoising (NeurIPS 2026 · similarity 0.80)
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.