Outlier Path: A Homotopy Algorithm for Robust SVM

Authors: Shinya Suzumura, Kohei Ogawa, Masashi Sugiyama, Ichiro Takeuchi

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we compared the proposed outlier-path (OP) algorithm with the concave-convex procedure (CCCP) (Yuille & Rangarajan, 2002). In most of the existing RSVM studies, CCCP or a variant called difference of convex (DC) programming are used for optimizing RSVM (Shen et al., 2003; Krause & Singer, 2004; Liu et al., 2005; Liu & Shen, 2006; Collobert et al., 2006; Wu & Liu, 2007). We used the 10 benchmark data sets listed in Table 1. We randomly divided each data set into the training (40%), validation (30%), and test (30%) sets for training, model selection (including the selection of θ or s), and performance evaluation, respectively.
Researcher Affiliation Academia Shinya SUZUMURA suzumura.mllab.nit@gmail.com Nagoya Institute of Technology Gokiso-cho, Showa-ku, Nagoya, Aichi 466 8555 Japan Kohei OGAWA ogawa.mllab.nit@gmail.com Nagoya Institute of Technology Gokiso-cho, Showa-ku, Nagoya, Aichi 466 8555 Japan Masashi Sugiyama sugi@cs.titech.ac.jp Tokyo Institute of Technology O-okayama, Meguro-ku, Tokyo 152-8552, Japan Ichiro Takeuchi takeuchi.ichiro@nitech.ac.jp Nagoya Institute of Technology Gokiso-cho, Showa-ku, Nagoya, Aichi 466 8555 Japan
Pseudocode Yes Algorithm 1 Outlier Path Algorithm Algorithm 2 Continuous Step (C-step) Algorithm 3 Discontinuous Step (D-step)
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We used the 10 benchmark data sets listed in Table 1. ... Table 1. Benchmark data sets. n and d denote the number of instances and the input dimensionality, respectively. Data D1 Breast Cancer Diagnostic ... D2 Australian Credit Approval ... D5 Spambase ... D7 Gisette ...
Dataset Splits Yes We randomly divided each data set into the training (40%), validation (30%), and test (30%) sets for training, model selection (including the selection of θ or s), and performance evaluation, respectively.
Hardware Specification No The paper discusses computational time and presents results in Figure 5, but it does not specify any hardware details (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies, libraries, or solvers used in their implementation.
Experiment Setup Yes In all the setups, the regularization parameter C was chosen from {0.01, 0.1, 1, 10, 100}, while the candidates of the homotopy parameter θ or s were set as follows: In OP-θ, all the break-points θBP were considered as the candidates (note that the local solutions at each break-point have been already computed in the homotopy computation). In OP-s, all the break-points for s BP between sinit := mini Nn yi f(xi) and 0 are considered as the candidates.