Generalised f-Mean Aggregation for Graph Neural Networks
Authors: Ryan Kortvelesy, Steven Morad, Amanda Prorok
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper, we run three experiments. First, we show that Gen Agg can perform regression to recover any standard aggregation function. Then, we evaluate Gen Agg and several baselines inside of a GNN. The resulting GNN architectures are given the same task of regressing upon graph-structured data generated with a standard aggregator. This tests if it is possible for a GNN with a given aggregator to represent data which was generated by different underlying aggregators. Finally, we provide practical results by running experiments on public GNN benchmark datasets: CLUSTER, PATTERN, CIFAR10, and MNIST [6]. |
| Researcher Affiliation | Academia | Ryan Kortvelesy, Steven Morad and Amanda Prorok University of Cambridge {rk627, sm2558, asp45}@cam.ac.uk |
| Pseudocode | No | No structured pseudocode or algorithm blocks are explicitly present in the paper. |
| Open Source Code | Yes | Our code can be found at: https://github.com/Acciorocketships/generalised-aggregation. |
| Open Datasets | Yes | Finally, we provide practical results by running experiments on public GNN benchmark datasets: CLUSTER, PATTERN, CIFAR10, and MNIST [6]. |
| Dataset Splits | No | The paper uses standard benchmark datasets but does not explicitly state the specific train/validation/test splits (e.g., percentages or counts) used for reproduction in its main text. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments are mentioned in the paper. |
| Software Dependencies | No | The paper mentions 'Py Torch Geometric' but does not provide specific version numbers for it or any other software dependencies. |
| Experiment Setup | Yes | The GNN is implemented with 4 layers of Graph Conv [16] with Mish activation (after every layer except the last), where the default aggregation function is replaced by a parametrised aggregator Lθ: ... a 4-layer Graph Conv [16] GNN with a hidden size of 64 |