Instance Specific Metric Subspace Learning: A Bayesian Approach

Authors: Han-Jia Ye, De-Chuan Zhan, Yuan Jiang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiment on synthetic data shows that the learned results are with good interpretability. Moreover, comprehensive results on real world datasets validate the effectiveness and robustness of ISMETS. In this section, we first illustrate the mechanism of ISMETS with synthetic data, and then compare ISMETS with other distance metric learning methods. In order to investigate the abilities of handling unlabeled data and the robustness of ISMETS, we conduct more experiments to discover the influence on label ratio and the number of metric bases.
Researcher Affiliation Academia Han-Jia Ye and De-Chuan Zhan and Yuan Jiang National Key Laboratory for Novel Software Technology, Nanjing University Nanjing, 210023, China {yehj, zhandc, jiangy}@lamda.nju.edu.cn
Pseudocode No The paper describes the model and inference mathematically but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any information or link regarding open-source code availability.
Open Datasets Yes We test ISMETS on 8 UCI datasets and 4 real Bioinformatic datasets (GDS3286, GSE4115, GDS2771, GDS531). Can be open accessed from www.ncbi.nlm.nih.gov/geo/
Dataset Splits No In each trial, we randomly split the data into training set (67%) and test set (33%). In the training set, 30% data are labeled examples and the remains are unlabeled. The paper specifies training and test splits but does not explicitly mention a separate validation set.
Hardware Specification Yes Moreover, experiments running on computational servers with 2.66GHz 2 cores and 4GB memory show that ISMETS can be trained faster than some other ISM type methods.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes In our implementation, k is configured as 5. We run each method 30 random trials per data. The maximum training iteration for ISMETSt/i is fixed as 20. We use non-informative hyper-parameters in the training process (G onen and Margolin 2014; Zhao et al. 2014), i.e., we set ασ, βσ and elements in απ all as 1.