Geometric Scattering for Graph Data Analysis
Authors: Feng Gao, Guy Wolf, Matthew Hirn
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the application of our geometric scattering features in graph classification of social network data, and in data exploration of biochemistry data. ... Our results in Sec. 4.1 show that on social network data, geometric scattering features enable classic RBF-kernel SVM to match, if not outperform, leading graph kernel methods as well as most geometric deep learning ones. These experiments are augmented by additional results in Sec. 4.2 that show the geometric scattering SVM classification rate degrades only slightly when trained on far fewer graphs than is traditionally used in graph classification tasks. On biochemistry data, where graphs represent molecular structures of compounds (e.g., Enzymes or proteins), we show in Sec. 4.3 that scattering features enable significant dimensionality reduction. Finally, to establish their descriptive qualities, in Sec. 4.4 we use geometric scattering features extracted from enzyme data (Borgwardt et al., 2005) to infer emergent patterns of enzyme commission (EC) exchange preferences in enzyme evolution, validated with established knowledge from Cuesta et al. (2015). |
| Researcher Affiliation | Academia | 1Department of Computational Math., Science and Engineering, Michigan State University, East Lansing, MI, USA; 2Department of Plant, Soil & Microbial Sciences, Michigan State University, East Lansing, MI, USA; 3Department of Mathematics and Statistics, Universit e de Montr eal, Montreal, QC, Canada; 4Department of Mathematics, Michigan State University, East Lansing, MI, USA. |
| Pseudocode | No | The paper describes the method mathematically and provides a visual representation in Figure 2(a), but it does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide a direct link or statement confirming the availability of its own source code for the methodology described. |
| Open Datasets | Yes | social network data taken from Yanardag & Vishwanathan (2015). ... The social network data provided by Yanardag & Vishwanathan (2015) contains graph structures... ... On biochemistry data, where graphs represent molecular structures of compounds (e.g., Enzymes or proteins)... ... ENZYMES dataset introduced in Borgwardt et al. (2005) |
| Dataset Splits | Yes | We evaluate the classification results of our SVM-based geometric scattering classification (GS-SVM) using ten-fold cross validation (explained in the supplement)... We performed graph classification under four training/validation/test splits: 80%/10%/10%, 70%/10%/20%, 40%/10%/50% and 20%/10%/70%. We did 10-fold, 5-fold and 2-fold cross validation for the first three splits. For the last split, we randomly formed a 10 folds pool, from which we randomly selected 3 folds for training/validation and repeated this process ten times. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU, GPU models) used for running the experiments. |
| Software Dependencies | No | The paper mentions using "Py GSP (2018)" and "standard SVM classifier with an RBF kernel" but does not provide specific version numbers for these software components or libraries. |
| Experiment Setup | Yes | We use the normalized scattering moments for classification, since they perform slightly better than the un-normalized moments. Also we use J = 5 and q = 4 for all scattering feature generations. ... In particular, we use here the standard SVM classifier with an RBF kernel, which is popular and effective in many applications and also performs well in this case. |