Fast SVM Trained by Divide-and-Conquer Anchors

Authors: Meng Liu, Chang Xu, Chao Xu, Dacheng Tao

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on multiple datasets demonstrate that our DCASVM is faster than the state-of-the-art algorithms without notably decreasing the accuracy of classification results.
Researcher Affiliation Academia Meng Liu , Chang Xu , Chao Xu , Dacheng Tao Key Laboratory of Machine Perception (MOE), Cooperative Medianet Innovation Center, School of Electronics Engineering and Computer Science, Peking University, China UBTech Sydney AI Institute, The School of Information Technologies, The University of Sydney mengliu@pku.edu.cn, c.xu@sydney.edu.au, xuchao@cis.pku.edu.cn, dacheng.tao@sydney.edu.au
Pseudocode Yes Algorithm 1 Approximate SVM trained on divide-and-conquer anchors, where the anchor number k is determined automatically. Input: training data X, sub-problem number s, random vector number d. Output: Parameters of DCA-SVM for classification Split training set into X+ and X according to their labels; (1) Divide and Conquer Step: for i = 1 to s do generate random projection matrix B; obtain anchors Ai of X(Bi)T by SNMF in Eq. (5). end for combing anchors to get X+ A and X A by Eq. (6); (2) Coefficients Learning Step: Determining the weight matrices F and noise matrices N of X = FX A + N . (3) Training procedure: train SVM using anchors X+ A and X A according to Eq. (7).
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the described methodology.
Open Datasets Yes We perform all compared algorithms on three real-world datasets: KDD99Lite, UCI Forest1 and IJCNN12. ... 1https://archive.ics.uci.edu/ml/datasets/Covertype 2http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/ binary.html#ijcnn1 3http://archive.ics.uci.edu/ml/datasets/KDD+Cup+1999+Data
Dataset Splits Yes For the sake of accuracy of the experiment, we partitioned the data randomly for five-fold cross-validation.
Hardware Specification No The paper does not specify the hardware used for running the experiments (e.g., CPU, GPU models, memory).
Software Dependencies No The paper mentions software like LIBSVM and SVMlight, and SVMperf, but does not provide specific version numbers for these or other software dependencies used in their experiments.
Experiment Setup Yes The parameter C varies in the range {2 6, 2 5, . . . , 25, 26}. Our proposed DCA-SVM will be compared with AESVM... We set the parameter ϵ = 10 2 when using AESVM [Nandan et al., 2014]. ... SVMperf... We set the given number of support vectors as 1000 in our experiments.