Task-Robust Model-Agnostic Meta-Learning

Authors: Liam Collins, Aryan Mokhtari, Sanjay Shakkottai

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

Reproducibility Variable Result LLM Response
Research Type Experimental We also provide an upper bound on the new task generalization error that captures the advantage of minimizing the worst-case task loss, and demonstrate this advantage in sinusoid regression and image classification experiments.
Researcher Affiliation Academia Liam Collins ECE Department University of Texas at Austin Austin, TX 78712 liamc@utexas.edu Aryan Mokhtari ECE Department University of Texas at Austin Austin, TX 78712 mokhtari@austin.utexas.edu Sanjay Shakkottai ECE Department University of Texas at Austin Austin, TX 78712 sanjay.shakkottai@utexas.edu
Pseudocode Yes Algorithm 1 Task-Robust MAML (TR-MAML)
Open Source Code Yes 1The code for TR-MAML is available at: https://github.com/lgcollins/tr-maml.
Open Datasets Yes We experiment in this setting using the Omniglot [19] and mini-Image Net [37] datasets.
Dataset Splits Yes We use the same (meta-) train/validation/test splits as in [36].
Hardware Specification No The paper does not specify any hardware details such as GPU models, CPU types, or memory amounts used for the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Both algorithms use one SGD step as the inner learning algorithm, and the same fully-connected network architecture as in [12] for the learning model. ... We meta-train for 60,000 iterations with a batch size of 2 task instances, and 5 steps of gradient descent for local adaptation.