Self-Supervised Set Representation Learning for Unsupervised Meta-Learning

Authors: Dong Bok Lee, Seanie Lee, Kenji Kawaguchi, Yunji Kim, Jihwan Bang, Jung-Woo Ha, Sung Ju Hwang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We also empirically validate its effectiveness on various benchmark datasets, showing that Set-Sim CLR largely outperforms both UML and instance-level self-supervised learning baselines.
Researcher Affiliation Collaboration Dong Bok Lee1 Seanie Lee1 Kenji Kawaguchi2 Yunji Kim3 Jihwan Bang3 Jung-Woo Ha3 Sung Ju Hwang1 KAIST1, National University of Singapore2, NAVER3 {markhi, lsnfamily02, sjhwang82}@kaist.ac.kr {yunji.kim, jihwan.bang, jungwoo.ha}@navercorp.com kenji@comp.nus.edu.sg
Pseudocode Yes We provide the pseudo-code for Set-Sim CLR described in Section 3.3. Algorithm 1 Meta-Training for Set-Sim CLR. Algorithm 2 Meta-Test for Set-Sim CLR.
Open Source Code Yes In Supplementary File, we further provide the code for reproducing the main experimental results in Table 1 and Figure 2.
Open Datasets Yes Dataset We use the Mini-Image Net dataset introduced by Ravi & Larochelle (2017)... Tiny-Image Net (Le & Yang, 2015), CIFAR100 (Krizhevsky et al., 2009), Aircraft (Maji et al., 2013), Stanford Cars (Krause et al., 2013) and CUB (Wah et al., 2011) datasets.
Dataset Splits Yes We use 64 classes for unsupervised meta-training, 16 classes for meta-validation, and the remaining 20 classes for meta-test.
Hardware Specification No The paper mentions running augmentations 'on GPU' but does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for experiments.
Software Dependencies No The paper mentions software packages like 'scikit-learn', 'Kornia framework', and 'huggingface transformers library' but does not specify their version numbers, which is required for reproducibility.
Experiment Setup Yes We optimize the base encoder, set encoder and head network for 400 epochs using Adam optimizer (Kingma & Ba, 2015) with default settings (i.e., β1 = 0.9 and β2 = 0.999). We use constant learning rate of 0.001. For our method Set-Sim CLR, we apply the augmentations (which is defined in Appendix J) 8 times to the mini-batch of 64 images (i.e., M = 64, V = 8), resulting in 4 elements in each set, while performing the same augmentation twice on the mini-batch of 256 images (i.e., M = 256, V = 2) for the other baselines.