Disentangled Partial Label Learning

Authors: Wei-Xuan Bao, Yong Rui, Min-Ling Zhang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments over various datasets demonstrate that our approach outperforms the stateof-the-art counterparts.
Researcher Affiliation Collaboration Wei-Xuan Bao1, 2, Yong Rui3, Min-Ling Zhang1, 2* 1School of Computer Science and Engineering, Southeast University, Nanjing, China 2Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China 3Lenovo Research, Lenovo Group Ltd., Beijing, China baowx@seu.edu.cn, yongrui@lenovo.com, zhangml@seu.edu.cn
Pseudocode Yes The complete procedure of TERIAL is summarized in Appendix A.1.
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described in the paper.
Open Datasets Yes Five popular benchmark datasets are employed to generate synthetic PL data sets, including MNIST (Le Cun et al. 1998), Kuzushiji-MNIST(abbreviated as KMNIST) (Clanuwat et al. 2018), Fashion-MNIST(abbreviated as FMNIST) (Xiao, Rasul, and Vollgraf 2017), SVHN (Netzer et al. 2011) and CIFAR-10 (Krizhevsky, Hinton et al. 2009).
Dataset Splits Yes Their hyper-parameters are specified according to the suggested parameter settings or searched to maximize the accuracy on a validation set containing 10% of the training samples.
Hardware Specification No The paper mentions using "DNN based methods" and models like "Res Net", which implies GPU usage, but it does not provide any specific details about the CPU or GPU models, or other hardware specifications used for the experiments.
Software Dependencies No The paper mentions the use of "MLP", "Res Net", and "stochastic gradient descent (SGD)" but does not provide specific version numbers for any software libraries or frameworks (e.g., PyTorch, TensorFlow, Python version).
Experiment Setup Yes For TERIAL, the assumed number of latent factors is set as K = 10 on datasets of MNIST, KMNIST, FMNIST and K = 8 on SVHN and CIFAR-10. The number of disentangling layers is set as L = 2, which is sufficient to achieve state-of-the-art performance for our proposed approach. We search the initial learning rate from {10 1, 10 2, 10 3, 10 4} and the weight decay from {10 2, 10 3, 10 4, 10 5}. The mini-batch size is set as 256 and the number of epochs is set as 200. All the models are trained with stochastic gradient descent (SGD) (Robbins and Monro 1951) optimizer with momentum 0.9. In this paper, the maximum number of iterations is set to be T = 6, which suffices to yield stable performance for the proposed approach. the balancing factor is set as α = 0.6 in this paper.