Semi-supervised Learning with Support Isolation by Small-Paced Self-Training

Authors: Zheng Xie, Hui Sun, Ming Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on both benchmark and pneumonia diagnosis tasks show that our method is effective.
Researcher Affiliation Academia Zheng Xie, Hui Sun, Ming Li National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China {xiez,sunh,lim}@lamda.nju.edu.cn
Pseudocode Yes Algorithm 1: Small-Paced Self-Training Framework ... Algorithm 2: Small-Paced Self-Training Algorithm
Open Source Code No The paper does not provide any statements about releasing source code or links to a code repository.
Open Datasets Yes We compare the methods on commonly used CIFAR10, CIFAR100 (Krizhevsky 2009) dataset and real-world X-ray pneumonia identification task (Kermany et al. 2018)
Dataset Splits No The paper mentions using well-known datasets but does not explicitly provide the training, validation, or test split percentages or sample counts used for its experiments.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper mentions using Res Net-50 as a backbone but does not specify any software dependencies with version numbers.
Experiment Setup Yes For linear base models f(x) = w x + b, we optimize ramp loss with ℓ2 regularization... For deep models, we use Large Margin Deep Networks (Elsayed et al. 2018) as the base models... Generally, for datasets with normalized features, we search δ in [0.1, 0.5]... All methods adopt Res Net-50 pre-trained on imagenet as the backbone... those predictions with large margin |f (t)(x U)| > θ will be accepted as pseudo-labeled data D(t) U. if | D(t) U | = 0 then decrease θ.