Bayesian Meta-Learning for the Few-Shot Setting via Deep Kernels

Authors: Massimiliano Patacchiola, Jack Turner, Elliot J. Crowley, Michael O'Boyle, Amos J. Storkey

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate that DKT outperforms several state-of-the-art algorithms in few-shot classification, and is the state of the art for cross-domain adaptation and regression.
Researcher Affiliation Academia Massimiliano Patacchiola School of Informatics University of Edinburgh mpatacch@ed.ac.uk Jack Turner School of Informatics University of Edinburgh jack.turner@ed.ac.uk Elliot J. Crowley School of Engineering University of Edinburgh elliot.j.crowley@ed.ac.uk Michael O Boyle School of Informatics University of Edinburgh mob@inf.ed.ac.uk Amos Storkey School of Informatics University of Edinburgh a.storkey@ed.ac.uk
Pseudocode Yes Algorithm 1 Deep Kernel Transfer (DKT) in the few-shot setting, train and test functions.
Open Source Code Yes We derive two versions of DKT for both the regression and the classification setting, comparing it against recent methods on a standardized benchmark environment; the code is released with an open-source license1. 1https://github.com/Bayes Watch/deep-kernel-transfer
Open Datasets Yes We consider two challenging datasets: the Caltech-UCSD Birds (CUB-200, Wah et al., 2011), and mini-Image Net (Ravi and Larochelle, 2017).
Dataset Splits Yes In the most common scenario, training, validation and test datasets each consist of distinct tasks sampled from the same overall distribution over tasks.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments, only discussing the software framework used.
Software Dependencies No The paper mentions 'Py Torch and GPy Torch' but does not provide specific version numbers for these or any other software components, which are necessary for reproducibility.
Experiment Setup Yes The training set is composed of 5 support and 5 query points, and the test set of 5 support and 200 query points. All the experiments are 5-way (5 random classes) with 1 or 5-shot (1 or 5 samples per class in the support set). A total of 16 samples per class are provided for the query set.