Bayesian Semi-supervised Learning with Graph Gaussian Processes

Authors: Yin Cheng Ng, Nicolò Colombo, Ricardo Silva

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed model shows extremely competitive performance when compared to the state-of-the-art graph neural networks on semi-supervised learning benchmark experiments, and outperforms the neural networks in active learning experiments where labels are scarce.
Researcher Affiliation Academia Yin Cheng Ng1, Nicolò Colombo1, Ricardo Silva1,2 1Statistical Science, University College London 2The Alan Turing Institute {y.ng.12, nicolo.colombo, ricardo.silva}@ucl.ac.uk
Pseudocode No The paper describes the proposed model and its components in detail but does not include any explicit pseudocode or algorithm blocks.
Open Source Code No Implementing the GGP with its variational inference algorithm amounts to implementing a new kernel function that follows Equation 12 in the GPflow Python package.1 1https://github.com/markvdw/GPflow-inter-domain. The link points to a general GPflow extension project, not specific open-source code for the GGP model described in the paper.
Open Datasets Yes The three benchmark data sets, as described in Table 2, are citation networks with bag-of-words (BOW) features, and the prediction targets are the topics of the scientific papers in the citation networks.
Dataset Splits Yes The semi-supervised classification experiments in this section exactly replicate the experimental setup in [25], where the GCN is known to perform well. (This references [25] which specifies validation splits). Furthermore, the model does not require a validation data set for early stopping to control over-fitting.
Hardware Specification No The paper does not provide any specific details regarding the hardware used to run the experiments, such as GPU/CPU models, memory, or cloud instance types.
Software Dependencies No The paper mentions 'GPflow Python package' and 'ADAM optimizer' but does not provide specific version numbers for these or any other software dependencies, such as Python version or library versions.
Experiment Setup Yes The GGP base kernel of choice is the 3rd degree polynomial kernel... We re-weighed the BOW features using the popular term frequency-inverse document frequency (TFIDF) technique [40]. The variational parameters and the hyper-parameters were jointly optimized using the ADAM optimizer [24].