Who Is Your Right Mixup Partner in Positive and Unlabeled Learning

Authors: Changchun Li, Ximing Li, Lei Feng, Jihong Ouyang

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experimental results demonstrate the effectiveness of the heuristic mixup technique in PU learning and show that P3Mix can consistently outperform the state-of-the-art PU learning methods.
Researcher Affiliation Academia 1College of Computer Science and Technology, Jilin University, China 2College of Computer Science, Chongqing University, China 3Imperfect Information Learning Team, RIKEN Center for Advanced Intelligence Project, Japan
Pseudocode Yes Algorithm 1 Training procedure of P3Mix, P3Mix-E and P3Mix-C
Open Source Code No The paper provides links to the code for various baseline methods (e.g., nn PU, Self-PU, PAN, VPU, MIXPUL) in the appendix, but it does not provide an explicit link or statement about the availability of the source code for their proposed P3Mix method.
Open Datasets Yes In the experiments, we employ three prevalent benchmark datasets, including Fashion MNIST (F-MNIST) (Xiao et al., 2017),3 CIFAR-10 (Krizhevsky, 2016),4 and STL-10 (Coates et al., 2011).5
Dataset Splits Yes For each dataset, we randomly select 1,000 positive instances from the training set, and 500 instances as the validation set.
Hardware Specification No The paper mentions implementing P3Mix using Pytorch and Adam algorithm but does not specify any hardware details like GPU models, CPU types, or cloud computing resources used for experiments.
Software Dependencies No The paper states: 'We implement P3Mix, P3Mix-E and P3Mix-C by using Pytorch (Paszke et al., 2019) with the Adam algorithm (Kingma & Ba, 2014).' While PyTorch is mentioned with a citation year, a specific version number like 'PyTorch 1.9' is not provided. Adam is an algorithm, not a software dependency with a version.
Experiment Setup Yes We employ the cross entropy function as the loss function ℓof Eq.(2), fix the mixup hyperparameter α to 1 and the size k of the candidate mixup pool Xcnd to 100, and choose the coefficient parameter β from {0.8, 0.9, 1.0}, the thresholding parameter γ from {0.85, 0.9, 0.95}. ... Specially, the early-learning regularization parameter of P3Mix-E is chosen from {1.0, 2.0, 3.0, 4.0, 5.0}.