Learning to Balance: Bayesian Meta-Learning for Imbalanced and Out-of-distribution Tasks
Authors: Hae Beom Lee, Hayeon Lee, Donghyun Na, Saehoon Kim, Minseop Park, Eunho Yang, Sung Ju Hwang
ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our Bayesian Task-Adaptive Meta-Learning (Bayesian TAML) on multiple realistic taskand class-imbalanced datasets, on which it significantly outperforms existing meta-learning approaches. Further ablation study confirms the effectiveness of each balancing component and the Bayesian learning framework. 5 EXPERIMENTS |
| Researcher Affiliation | Collaboration | KAIST1, Tmax Data2, AITRICS3, South Korea {haebeom.lee, hayeon926, eunhoy, sjhwang82}@kaist.ac.kr donghyun_na@tmax.co.kr, {shkim, mike_seop}@aitrics.com |
| Pseudocode | No | The paper describes the proposed model and inference process using mathematical equations and descriptive text, but it does not include a clearly labeled "Pseudocode" or "Algorithm" block. |
| Open Source Code | Yes | Code is available at https://github.com/haebeom-lee/l2b |
| Open Datasets | Yes | Datasets We validate our method on the following benchmark datasets. CIFAR-FS: This dataset (Bertinetto et al., 2019) is a variant of CIFAR-100 dataset... mini Image Net: This dataset (Vinyals et al., 2016) is a subset of the Image Net dataset... SVHN: This dataset (Netzer et al., 2011) is frequently used as an OOD dataset... |
| Dataset Splits | Yes | We split the dataset into 64/16/20 classes for training/validation/test. Aircraft: We split this dataset (Maji et al., 2013) into 70/15/15 classes for metatraining/validation/test with 100 examples for each class. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments (e.g., CPU/GPU models, memory, etc.). |
| Software Dependencies | No | The paper mentions using PyTorch for implementation but does not specify its version number or any other software dependencies with their respective versions. |
| Experiment Setup | Yes | We set the number of inner-gradient steps to 5 for meta-training and 10 for meta-testing, for all the models that take inner-gradient steps. We meta-train all models for total 50K iterations with meta-batch size set to 4. The outer learning rate is set to 0.001 for all the baselines and our models. |