Multiclass Capped _p-Norm SVM for Robust Classifications
Authors: Feiping Nie, Xiaoqian Wang, Heng Huang
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present experimental results showing that employing the new capped ℓp-norm SVM method can consistently improve the classification performance, especially in the cases when the data noise level increases. The validation experiments have been conducted on six benchmark datasets. All empirical results demonstrate that our new capped ℓp-norm SVM method is robust to data outliers and consistently improve the classification performance. |
| Researcher Affiliation | Academia | Feiping Nie,1,2 Xiaoqian Wang,2 Heng Huang2 1School of Computer Science, OPTIMAL, Northwestern Polytechnical University, Xian 710072, Shaanxi, P. R. China 2Department of Computer Science and Engineering, University of Texas at Arlington, USA |
| Pseudocode | Yes | Algorithm 1 Re-weighted method to solve problem (13). Algorithm 2 Algorithm to solve the problem (12). |
| Open Source Code | No | The paper mentions using LIBSVM for comparison but does not state that the code for *their* proposed method (Capped SVM) is open-source or provide any access links to it. |
| Open Datasets | Yes | The six benchmark datasets involved in our experiments are: ALLAML data set (Fodor 1997), the Human Lung Carcinomas (LUNG) data set (Bhattacharjee et al. 2001), the Human Carcinomas (Carcinomas) data set (Su et al. 2001), the Prostate Cancer Gene Expression (Prostate-GE) data set (Singh et al. 2002), the Japanese Female Facial Expression (JAFFE) data set (Lyons, Kamachi, and Gyoba 1997), the chemical analysis of wine (Wine) data set, and the physical measurements of abalone (Abalone) data set, the first four of which are gene expression microarray data sets, the latter one is an image data set while the last one is a daily life data set from the UCI Machine Learning Repository (Bache and Lichman 2013). |
| Dataset Splits | Yes | Before classification, all data sets are normalized to the range of [0, 1] and randomly divided using 5-fold cross validation. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. It only discusses the methods and datasets. |
| Software Dependencies | No | The paper mentions exploiting "the program from LIBSVM" but does not specify a version number for LIBSVM or any other software dependencies with version numbers, making the software environment not reproducible. |
| Experiment Setup | Yes | For all other methods involving a parameter, including SVM, LSSVM and Capped SVM, we tuned the parameter to be {10 3, 10 2, 10 1, 1, 10, 100} separately and recorded the best results. In our method, we set the value of ε in a heuristic way, that is, in the first five iterations, we selected 10% data with the largest noise to determine ε. |