Optimality of Message-Passing Architectures for Sparse Graphs
Authors: Aseem Baranwal, Kimon Fountoulakis, Aukosh Jagannath
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We now demonstrate our results through experiments using pytorch and pytorch-geometric [Fey and Lenssen, 2019]. The following simulations are for the setting n = 10000 and d = 4 for binary classification on the CSBM. We implement Architecture 1 for the binary case, and perform full-batch training on a graph sampled from the CSBM with certain signals (mentioned in the figures), followed by an evaluation of the architecture on a new graph sampled from the same distribution. |
| Researcher Affiliation | Academia | Aseem Baranwal David R. Cheriton School of Computer Science University of Waterloo, Waterloo, Canada aseem.baranwal@uwaterloo.ca; Kimon Fountoulakis David R. Cheriton School of Computer Science University of Waterloo, Waterloo, Canada kimon.fountoulakis@uwaterloo.ca; Aukosh Jagannath Department of Statistics and Actuarial Science, Department of Applied Mathematics, David R. Cheriton School of Computer Science University of Waterloo, Waterloo, Canada a.jagannath@uwaterloo.ca |
| Pseudocode | Yes | Architecture 1. Given input data (A, X) where A {0, 1}n n is the adjacency matrix and X Rn d is the node feature matrix, define: H(0) = X, H(l) = σl(H(l 1)W(l) + 1nb(l)) for l [L], Q = sigmoid(Z), M(k) u,i = max j [C] n H(L) u,j + log(Qk i,j) o for k [ℓ], u [n], i [C]. Then the predicted label is given by ˆy = {ˆyu}u [n], where ˆyu = argmax i [C] Pk=1 A(k) u,: M(k) :,i |
| Open Source Code | No | The paper mentions using 'pytorch and pytorch-geometric' for experiments but does not provide concrete access information or a statement about open-sourcing its own developed code for the described methodology. |
| Open Datasets | No | The paper uses a synthetic data model, the 'Contextual Stochastic Block Model (CSBM)', generated for analysis. It describes the data generation process in Section 3.2 but does not provide concrete access information (link, DOI, repository) for a publicly available or open dataset. |
| Dataset Splits | No | The paper mentions 'full-batch training' and evaluating 'on a new graph sampled from the same distribution', which implies a test set, but does not provide specific details on train/validation/test splits, percentages, or a methodology for data partitioning. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions using 'pytorch and pytorch-geometric' but does not provide specific version numbers for these or any other software components. |
| Experiment Setup | No | The paper states 'perform full-batch training' and mentions 'gradient descent (SGD and Adam implementations)', but it does not provide specific hyperparameter values like learning rate, batch size, number of epochs, or other detailed training configurations for its own architecture. |