Multi-Class Learning using Unlabeled Samples: Theory and Algorithm

Authors: Jian Li, Yong Liu, Rong Yin, Weiping Wang

IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Coinciding with the theoretical analysis, experimental results demonstrate that the stated approach achieves better performance.
Researcher Affiliation Academia 1Institute of Information Engineering, Chinese Academy of Sciences 2School of Cyber Security, University of Chinese Academy of Sciences {lijian9026, liuyong, yinrong, wangweiping}@iie.ac.cn
Pseudocode Yes Algorithm 1 Proximal Stochastic Sub-gradient Singular Value Thresholding (PS3VT)
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of its methodology's code.
Open Datasets Yes We run PS3VT and the compared methods on 15 multi-class datasets and report the results in Table 3. Labeled and unlabeled samples are given by stratified random sampling from train data that 30% as labeled samples and the rest as unlabeled ones.
Dataset Splits Yes For fair comparison, before a method runs on any dataset, we employ 5-folds cross validation to obtain the optimal parameter set by grid search over candidate sets complexity parameter τA {10 15, 10 14, , 10 6}, unlabeled samples parameter τI {0, 10 15, 10 14, , 10 6}, local Rademacher complexity parameter τS {0, 10 10, 10 9, , 10 1}, step size 1 µ {101, 102, , 105} and tail parameter θ {0.5, 0.6, , 0.9} min(|K|, |d|).
Hardware Specification No The paper does not provide specific details regarding the hardware used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments.
Experiment Setup Yes For fair comparison, before a method runs on any dataset, we employ 5-folds cross validation to obtain the optimal parameter set by grid search over candidate sets complexity parameter τA {10 15, 10 14, , 10 6}, unlabeled samples parameter τI {0, 10 15, 10 14, , 10 6}, local Rademacher complexity parameter τS {0, 10 10, 10 9, , 10 1}, step size 1 µ {101, 102, , 105} and tail parameter θ {0.5, 0.6, , 0.9} min(|K|, |d|).