Efficient Data Point Pruning for One-Class SVM

Authors: Yasuhiro Fujiwara, Sekitoshi Kanai, Junya Arai, Yasutoshi Ida, Naonori Ueda3590-3597

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluated the efficiency and effectiveness of the proposed approach. We performed the experiments on three datasets of gisette, rcv1.binary, and real-sim downloaded from the website of LIBSVM.
Researcher Affiliation Collaboration NTT Software Innovation Center, 3-9-11 Midori-cho Musashino-shi, Tokyo, 180-8585, Japan NTT Communication Science Laboratories, 2-4 Seika-Cho Soraku-gun, Kyoto, Japan Osaka University, 1-5 Yamadaoka, Suita-shi, Osaka, Japan RIKEN Center for AIP, 1-4-1 Nihonbashi, Chuo-ku, Tokyo, 103-0027, Japan
Pseudocode Yes Algorithm 1 shows the approach to compute the initial setting of Lagrange multipliers. Algorithm 2 gives a full description of our approach, Quix.
Open Source Code No The paper does not provide any specific links or statements about the availability of its source code.
Open Datasets Yes We performed the experiments on three datasets of gisette, rcv1.binary, and real-sim downloaded from the website of LIBSVM.
Dataset Splits No The paper uses terms like "training time" and "training accuracy" but does not explicitly provide details on how the datasets were split into training, validation, and test sets, or specify explicit validation splits.
Hardware Specification Yes We conducted all experiments on a Linux server with 2.70 GHz Intel Xeon.
Software Dependencies No The paper mentions using "the original SMO (Platt 1998) as the solver" and the "RBF kernel function" but does not specify version numbers for SMO or any other software libraries or dependencies.
Experiment Setup Yes In the experiments, we set the number of landmark to 0.01 n for the Nystr om method. We set the number of features to 0.01 m for Random Fourier features. For the coherence criterion-based approach, we set threshold µ0 so that the number of support vectors is nν for parameter ν to ensure fair comparisons. We used the RBF kernel function and set kernel parameter γ = 1 m. Figure 1 shows the results with parameter ν values of 0.02, 0.04, and 0.06.