CMVAE: Causal Meta VAE for Unsupervised Meta-Learning

Authors: Guodong Qi, Huimin Yu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Results on toy datasets and three benchmark datasets demonstrate that our method can remove the context-bias and it outperforms other state-of-the-art unsupervised meta-learning algorithms because of bias-removal. Code is available at https://github.com/Guodong Qi/CMVAE.
Researcher Affiliation Academia Guodong Qi 1,2, Huimin Yu 1,2,3,4 1College of Information Science and Electronic Engineering, Zhejiang University 2ZJU-League Research & Development Center 3State Key Lab of CAD&CG, Zhejiang University 4Zhejiang Provincial Key Laboratory of Information Processing, Communication and Networking {guodong qi, yhm2005}@zju.edu.cn
Pseudocode Yes Algorithm 1: Unsupervised Causal Meta-training
Open Source Code Yes Code is available at https://github.com/Guodong Qi/CMVAE.
Open Datasets Yes Omniglot, mini Image Net, Celeb A. The paper references 'mini Image Net. It is a subset of Image Net (Russakovsky et al. 2015)...'
Dataset Splits Yes We take 1200, 100, 323 classes for training, validation and test, respectively. (Omniglot) ... we take 64 classes for training, 16 for validation and 20 for test, respectively. (mini Image Net)
Hardware Specification No The paper does not explicitly describe the hardware used for experiments, such as specific GPU or CPU models. It only mentions general training parameters and networks.
Software Dependencies No The paper mentions using 'Adam (Kingma and Ba 2015)' as an optimizer but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes The number of iterations for causal-EM steps of all experiment is 10. The hyper-parameters γ, λ1 and λ2 are chosen based on the validation accuracy. We train all models for 60,000 iterations using Adam (Kingma and Ba 2015).