Learning Latent Graph Structures and their Uncertainty
Authors: Alessandro Manenti, Daniele Zambon, Cesare Alippi
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results validate our theoretical claims and demonstrate the effectiveness of the proposed approach. ... 6. Experiments |
| Researcher Affiliation | Academia | 1Universit a della Svizzera italiana, IDSIA, Lugano, Switzerland 2Politecnico di Milano, Milan, Italy. |
| Pseudocode | No | The paper describes methods and algorithms but does not provide any explicitly labeled pseudocode or algorithm blocks with structured formatting. |
| Open Source Code | Yes | 1Code available at https://github.com/allemanenti/Learning-Calibrated-Structures |
| Open Datasets | Yes | To demonstrate that our method learns meaningful graph distributions in real-world settings, we train a neural network on air quality data in Beijing (Zheng et al., 2013). |
| Dataset Splits | Yes | We result in a dataset of 35k input-output pairs (x, y), 80% of which are used as training set, 10% as validation set, and the remaining 10% as test set. |
| Hardware Specification | Yes | The paper s experiments were run on a workstation with AMD EPYC 7513 processors and NVIDIA RTX A5000 GPUs; on average, a single model training terminates in a few minutes with a memory usage of about 1GB. |
| Software Dependencies | No | The developed code relies on Py Torch (Paszke et al., 2019) and the following additional open-source libraries: Py Torch Geometric (Fey & Lenssen, 2019), Num Py (Harris et al., 2020) and Matplotlib (Hunter, 2007). While these libraries are mentioned, specific version numbers (e.g., PyTorch 1.9) are not provided. |
| Experiment Setup | Yes | The model is trained using Adam optimizer (Kingma & Ba, 2014) with parameters β1 = 0.9, β2 = 0.99. Where not specified, the learning rate is set to 0.05 and decreased to 0.01 after 5 epochs. We grouped data points into batches of size 128. Initial values of θ are independently sampled from the U(0.0, 0.1) uniform distribution. |