Towards Safe Semi-Supervised Learning for Multivariate Performance Measures

Authors: Yu-Feng Li, James Kwok, Zhi-Hua Zhou

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that the proposed method can effectively improve the safeness of SSL under multiple multivariate performance measures.
Researcher Affiliation Academia 1 National Key Laboratory for Novel Software Technology, Nanjing University 2 Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing, 210023 3 Department of Computer Science & Engineering, Hong Kong University of Science and Technology, Hong Kong
Pseudocode Yes Algorithm 1 Cutting-plane algorithm for Eq.(7).
Open Source Code No The paper mentions a URL for S4VM (a baseline method) but does not provide a link or explicit statement about the open-source code for their proposed UMVP method.
Open Datasets Yes Downloaded from http://www.csie.ntu.edu.tw/~cjlin/ libsvmtools/datasets/, http://www.kyb.tuebingen.mpg.de/sslbook/, and from (Mallapragada et al. 2009) (for the ethn data set). For the MNIST data set, we focus on its four most difficult binary classification tasks (Zhang, Tsang, and Kwok 2007).
Dataset Splits No The paper specifies '1% of the samples are labeled and the rest are unlabeled' and that 'Each experiment is repeated 10 times, and the average performance on the unlabeled data is reported', but it does not define a distinct validation dataset split.
Hardware Specification Yes Experiments are run on a PC with a 3.2GHz Core2 Duo CPU and 4GB memory.
Software Dependencies Yes The experiments are used with MATLAB 8.0.1 and LIBLINEAR 1.91.
Experiment Setup Yes For all methods, the C parameter in SVM is set to 1 and the linear kernel is used. Parameters of S4VM are set as recommended in the package. ... ϵ in Algorithm 1 is set to 10 6. When Fβ is used as the performance measure, the c in Proposition 2 is set to the average number of positive samples for the multiple learners., i.e., 1 b b i=1(yi) 1.