Multi-Label Informed Feature Selection

Authors: Ling Jian, Jundong Li, Kai Shu, Huan Liu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical studies on real-world datasets demonstrate the effectiveness and efficiency of the proposed framework.
Researcher Affiliation Academia 1. Computer Science and Engineering, Arizona State University, Tempe, 85281, USA 2. College of Science, China University of Petroleum, Qingdao, 266555, China
Pseudocode Yes The pseudocode of the multi-label informed feature selection framework MIFS is illustrated in Algorithm 1.
Open Source Code No The paper does not provide a specific link or explicit statement about the release of its own source code for the methodology.
Open Datasets Yes Experiments are conducted on four publicly available benchmark datasets1, including one image dataset (i.e. Scene [Boutell et al., 2004]) and three text datasets from RCV1 [Lewis et al., 2004]. 1https://www.csie.ntu.edu.tw/ cjlin/libsvmtools/datasets/
Dataset Splits Yes To have a fair comparison with existing methods, we decompose the multi-labeled classification problem into multiple binary classification problems, and then employ SVM to learn these binary classifiers with a five-fold cross validation. Table 1: Details of four benchmark datasets. Dataset # Training # Test # Features # Labels
Hardware Specification No The paper discusses running time and computational efficiency but does not specify any hardware details (e.g., CPU, GPU models, memory) used for the experiments.
Software Dependencies No The paper mentions using 'Liblinear toolbox' but does not specify a version number for it or other software dependencies.
Experiment Setup Yes In MIFS, there are some parameters need to be set in advance. First, to model the local geometry structure in the input space X, we set the parameters σ2 and p as 1 and 5, respectively. There are three important regularization parameters λ, β and γ in MIFS. For all these methods, we report the best results of the optimal parameters in terms of classification performance. The experiments are repeated 5 times and averaged.