Online Constrained Meta-Learning: Provable Guarantees for Generalization

Authors: Siyuan Xu, Minghui Zhu

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Moreover, we provide a practical algorithm for the framework and validate its superior effectiveness through experiments conducted on meta-imitation learning and few-shot image classification.
Researcher Affiliation Academia Siyuan Xu & Minghui Zhu School of Electrical Engineering and Computer Science The Pennsylvania State University University Park, PA 16801 {spx5032, muz16}@psu.edu
Pseudocode Yes Algorithm 1 Online Constrained Meta-Learning Framework
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We test the algorithms on two few-shot learning datasets, CUB [53] and mini-Image Net [52].
Dataset Splits No The paper describes the number of data samples used for training and validation for each task (e.g., '|Dtr 0 | = 50', '|Dval 0 | = 50'), but it does not specify fixed training/validation/test splits (e.g., 80/10/10 percentages or specific counts) for the overall datasets that are needed for reproduction.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions software like PyTorch and optimizers like Adam, but it does not provide specific version numbers for these software components, which are necessary for reproducible descriptions.
Experiment Setup Yes In the experiments, the total number of tasks is T = 100. For each task, the number of training data |Dtr 0 | = 50, |Dtr + | = 50, and the validation data |Dval 0 | = 50. The regularization parameter λ = 0.1, and the perturbation parameter η = 0.01. ... For Few-shot image classification, the total number of tasks is T = 200. We consider 5-way 1-shot and 5-way 5-shot learning. The training data |Dtr 0 | = 5 and |Dtr + | = 5 for 5-shot learning, and |Dtr 0 | = 1 and |Dtr + | = 1 for 1-shot learning. The validation data |Dval 0 | = 50.