Latent Graph Inference using Product Manifolds
Authors: Haitz Sáez de Ocáriz Borde, Anees Kazi, Federico Barbero, Pietro Lio
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our novel approach is tested on a wide range of datasets, and outperforms the original d DGM model. |
| Researcher Affiliation | Academia | Haitz S aez de Oc ariz Borde University of Oxford Anees Kazi Harvard University Federico Barbero University of Oxford Pietro Li o University of Cambridge |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | To make this work accessible to the wider machine learning community, we have created a new Py Torch Geometric layer. |
| Open Datasets | Yes | We test our approach on 15 datasets which includes standard homophilic graph datasets, heterophilic graphs, large-scale graphs, molecular datasets, and datasets for other real-world applications such as brain imaging and aerospace engineering. In particular, we use the OGB-Arxiv and the OGBProducts datasets. We work with the QM9 (Ramakrishnan et al. (2014); Ruddigkeit et al. (2012)) and Alchemy (Morris et al. (2020)) datasets which are well known in the Geometric Deep Learning literature. We first focus on standard graph datasets widely discussed in the Geometric Deep Learning literature such as Cora, Cite Seer (Yang et al. (2016); Lu & Getoor (2003); Sen et al. (2008)), Pub Med, Physics and CS (Shchur et al. (2018)). |
| Dataset Splits | Yes | Table 2: Results for heterophilic and homophilic datasets combining GCN diffusion layers with the latent graph inference system. We apply a neighbor sampler to track message passing dependencies for the subsampled nodes. For graph subsampling, we sample up to 1,000 neighbours per node and use a batch size of 1,000. (OGB-Arxiv) For graph subsampling, we sample up to 200 neighbors per node and use a batch size of 1,000. (OGB-Products) |
| Hardware Specification | Yes | Most of the experiments were performed using NVIDIA Tesla T4 Tensor Core GPUs with 16 GB of GDDR6 memory, NVIDIA P100 GPUs with 16 GB of Co Wo S HBM2 memory, or NVIDIA Tesla K80 GPUs with 24 GB of GDDR5 memory. |
| Software Dependencies | No | The paper mentions using "Py Torch Geometric layer" and "Kernel Operations (Ke Ops)" but does not specify version numbers for these or other software components. |
| Experiment Setup | Yes | We apply a learning rate of lr 10 2 and a weight decay of wd 10 4. Models are trained for about 1,500 epochs. (Classical datasets) We apply lr 10 2 and wd 10 3, and we train the models for about 1,000 epochs. (Heterophilic datasets) We use lr 10 3, wd 0, and train for 100 epochs. For graph subsampling, we sample up to 1,000 neighbours per node and use a batch size of 1,000. (OGB-Arxiv) |