The Effect of Diversity in Meta-Learning

Authors: Ramnath Kumar, Tristan Deleu, Yoshua Bengio

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

Reproducibility Variable Result LLM Response
Research Type Experimental our experiments draw into question the efficacy of our learned models: similar manifolds can be learned with a subset of the data (lower task diversity). This finding questions the advantage of providing more data to the model, and (ii) adding diversity to the task distribution (higher task diversity) sometimes hinders the model and does not lead to a significant improvement in performance as previously believed. To strengthen our findings, we provide both empirical and theoretical evidence.
Researcher Affiliation Collaboration Ramnath Kumar1*, Tristan Deleu2, Yoshua Bengio2, 3 Google Research, India 1 Mila, Qu ebec Artificial Intelligence Institute, Universit e de Montr eal 2 CIFAR, IVADO 3
Pseudocode No The paper describes various samplers and methods but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Our source code is made available for additional reference 1 https://github.com/Ramnath Kumar181/Task-Diversity-metalearning
Open Datasets Yes To make an exhaustive study on the effect of task diversity in meta-learning, we train on four datasets: Omniglot (Lake et al. 2011), mini Imagenet (Ravi and Larochelle 2017), tiered Image Net (Ren et al. 2018), and Meta-Dataset (Triantafillou et al. 2019).
Dataset Splits No The paper describes train tasks and test tasks, and the use of support and query sets within episodic training, but does not explicitly detail a separate overall 'validation' dataset split for hyperparameter tuning.
Hardware Specification No The paper does not provide specific details regarding the hardware used for the experiments.
Software Dependencies No The paper does not provide specific version numbers for software dependencies such as libraries, frameworks, or programming languages.
Experiment Setup Yes More details about the models and their hyperparameters are discussed in Appendix B in (Kumar, Deleu, and Bengio 2022).