Learning Discrete Structures for Graph Neural Networks
Authors: Luca Franceschi, Mathias Niepert, Massimiliano Pontil, Xiao He
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct a series of experiments that analyze the behavior of the proposed method and demonstrate that it outperforms related methods by a significant margin. We conducted a series of experiments with three main objectives. The results on the incomplete graphs are shown in Figure 2 for Cora (left) and Citeseer (center). Table 1 lists the results for semi-supervised classification problems. |
| Researcher Affiliation | Collaboration | 1CSML, Istituto Italiano di Tecnologia, Genoa, Italy. 2University College London, London, UK. 3NEC Labs EU, Heidelberg, Germany. Work done in part as Luca Franceschi was visiting researcher at NEC. |
| Pseudocode | Yes | A sketch of the method is presented in Algorithm 1, while a more complete version that includes details on the hypergradient computation can be found in the appendix. |
| Open Source Code | Yes | LDS was implemented in Tensor Flow (Abadi et al., 2015) and is available at https://github.com/lucfra/LDS. |
| Open Datasets | Yes | Cora and Citeseer are two benchmark datasets that are commonly used to evaluate relational learners in general and GCNs in particular (Sen et al., 2008). In addition to Cora and Citeseer where we removed all edges, we evaluate LDS on benchmark datasets that are available in scikit-learn (Pedregosa et al., 2011) such as Wine, Breast Cancer (Cancer), Digits, and 20 Newsgroup (20news). We also use FMA, a dataset where 140 audio features are extracted from 7,994 music tracks and where the problem is genre classification (Defferrard et al., 2017). |
| Dataset Splits | Yes | We use the same dataset split and experimental setup of previous work (Yang et al., 2016; Kipf & Welling, 2017). We further split the validation set evenly to form the validation (A) and early stopping (B) sets. A fraction of the examples in the validation set is held-out to compute, in each outer iteration, the accuracy using the predictions of the empirically expected model (9). |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments (e.g., GPU models, CPU types, or memory specifications). |
| Software Dependencies | No | The paper mentions software like Tensor Flow and scikit-learn, but it does not specify any version numbers for these software components or other dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | We use the two layers GCN given by Eq. (2) with 16 hidden neurons and Re Lu activation. As additional regularization technique we apply dropout (Srivastava et al., 2014) with β = 0.5 as in previous work. We use Adam (Kingma & Ba, 2015) for optimizing L, tuning the learning rate γ from {0.005, 0.01, 0.02}. For LDS, we set the initial edge parameters θi,j to 0 except for the known edges (or those found by k NN) which we set to 1. We tune the step size η of the outer optimization loop and the number of updates τ used to compute the truncated hypergradient. Finally, we draw S = 16 samples to compute the output predictions (see Eq. (9)). For LDS and GCN, we apply early stopping with a window size of 20 steps. |