Convolutional Gaussian Processes

Authors: Mark van der Wilk, Carl Edward Rasmussen, James Hensman

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We investigate several variations of the convolutional kernel, and apply it to MNIST and CIFAR-10, where we obtain significant improvements over existing Gaussian process models. We also show how the marginal likelihood can be used to find an optimal weighting between convolutional and RBF kernels to further improve performance.
Researcher Affiliation Collaboration Mark van der Wilk Department of Engineering University of Cambridge, UK mv310@cam.ac.uk Carl Edward Rasmussen Department of Engineering University of Cambridge, UK cer54@cam.ac.uk James Hensman prowler.io Cambridge, UK james@prowler.io
Pseudocode No No pseudocode or algorithm blocks were found in the paper.
Open Source Code Yes Ours can be found on https://github.com/markvdw/convgp, together with code for replicating the experiments, and trained models.
Open Datasets Yes We investigate several variations of the convolutional kernel, and apply it to MNIST and CIFAR-10, where we obtain significant improvements over existing Gaussian process models.
Dataset Splits No No specific training/validation/test dataset split percentages or sample counts are provided in the paper for the used datasets (MNIST, CIFAR-10, Rectangles). It refers to 'training data' and 'test set' but not specific splits for reproducibility.
Hardware Specification No The paper mentions "leverage GPU implementations" and "memory constraints on the GPU" but does not specify any particular GPU model, CPU, or other hardware specifications used for experiments.
Software Dependencies No The paper states: "It is based on GPflow [30], allowing utilisation of GPUs." While GPflow is mentioned, no specific version number for GPflow or any other software dependency is provided.
Experiment Setup Yes We optimised using Adam [31] (0.01 learning rate & 100 data points per minibatch) and obtained 1.4% error and a negative log predictive probability (nlpp) of 0.055 on the test set. For CIFAR-10, it states: "All models use 1000 inducing inputs and are trained using Adam. Due to memory constraints on the GPU, a minibatch size of 40 had to be used for the weighted, additive and multi-channel models."