Online-Within-Online Meta-Learning

Authors: Giulia Denevi, Dimitris Stamos, Carlo Ciliberto, Massimiliano Pontil

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, preliminary numerical experiments confirm our theoretical findings.
Researcher Affiliation Academia Giulia Denevi1,2, Dimitris Stamos3, Carlo Ciliberto3,4 and Massimiliano Pontil1,3 1Istituto Italiano di Tecnologia (Italy), 2University of Genoa (Italy), 3University College of London (UK),4Imperial College of London (UK)
Pseudocode Yes Algorithm 1 Primal-dual online algorithm online Mirror Descent; Algorithm 2 Within-task algorithm; Algorithm 3 Meta-algorithm
Open Source Code Yes The code is available at https://github.com/dstamos/Adversarial-LTL
Open Datasets Yes 1) the Movielens-100k dataset3, containing the ratings of different users to different movies 2) the Mini-Wiki dataset from [3], containing sentences from Wikipedia pages and 3) the Jester-1 dataset4, containing the ratings of different users to different jokes. (Footnotes 3 and 4 provide URLs for Movielens-100k and Jester-1, and [3] is a formal citation for Mini-Wiki).
Dataset Splits Yes In all experiments, the hyper-parameters λ and were chosen by a meta-validation procedure (see App. I for more details) ... We split the available data into a training, validation and test set, with percentages 80%, 10% and 10% respectively.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes In all experiments, the hyper-parameters λ and were chosen by a meta-validation procedure (see App. I for more details) and we fixed 0 = I/d for the meta-algorithm in Alg. 8. (Appendix I further specifies: The inner algorithms were run with their default parameters, i.e. λ = 1 and n = 100 for the Movielens-100k dataset. The meta-parameters λ and were chosen by meta-validation, where λ was chosen from {0.1, 1, 10} and from {0.001, 0.01, 0.1, 1, 10}.)