Support Matrix Machines

Authors: Luo Luo, Yubo Xie, Zhihua Zhang, Wu-Jun Li

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on EEG and image classification data show that our model is more robust and efficient than the state-of-the-art methods.
Researcher Affiliation Academia Luo Luo RICKY@SJTU.EDU.CN Yubo Xie YUBOTSE@SJTU.EDU.CN Department of Computer Science and Engineering, Shanghai Jiao Tong University, Shanghai, China Zhihua Zhang ZHIHUA@SJTU.EDU.CN Institute of Data Science, Department of Computer Science and Engineering, Shanghai Jiao Tong University, China Wu-Jun Li LIWUJUN@NJU.EDU.CN National Key Laboratory for Novel Software Technology, Collaborative Innovation Center of Novel Software Technology and Industrialization, Department of Computer Science and Technology, Nanjing University, China
Pseudocode Yes Algorithm 1 ADMM for SMM Initialize S( 1) = b S(0) Rp q, Λ( 1) = bΛ(0) Rp q, ρ > 0, t(1) = 1, η (0, 1). for k = 1, 2, 3 . . . do (W(k), b(k)) = argmin (W,b) H(W, b) tr(bΛ(k)T W) + ρ 2||W b S(k)||2 F S(k) = argmin S G(S) + tr(bΛ(k)T S) + ρ 2||W(k) S||2 F Λ(k) = bΛ(k) ρ(W(k) S(k)) c(k) = ρ 1||Λ(k) bΛ(k)||2 F + ρ||S(k) b S(k)||2 F if c(k) < ηc(k 1) then t(k+1) = 1+ 2 b S(k+1) = S(k) + t(k) 1 t(k+1) (S(k) S(k 1)) bΛ(k+1) = Λ(k) + t(k) 1 t(k+1) (Λ(k) Λ(k 1)) else t(k+1) = 1 b S(k+1) = S(k 1) bΛ(k+1) = Λ(k 1) c(k) = η 1c(k 1) end if end for
Open Source Code Yes The code is available in http://bcmi.sjtu.edu.cn/ luoluo/code/smm.zip
Open Datasets Yes The EEG alcoholism data set arises to examine EEG correlates of genetic predisposition to alcoholism...http://kdd.ics.uci.edu/databases/eeg/ eeg.html; The EEG emotion data set (Zhu et al., 2014; Zheng et al., 2014); The student face data set contains 400 photos of Stanford University medical students (Nazir et al., 2010); The INRIA person data set...http://pascal.inrialpes.fr/data/human/
Dataset Splits Yes For each of the compared methods, we randomly sample 70% of the data set for training and the rest for testing. All the hyperparameters involved are selected via cross validation.
Hardware Specification Yes All experiments are implemented in Matlab R2011b on a workstation with Intel Xeon CPU X5675 3.06GHz (2 12 cores), 24GB RAM, and 64bit Windows Server 2008 system.
Software Dependencies Yes All experiments are implemented in Matlab R2011b
Experiment Setup Yes All the hyperparameters involved are selected via cross validation. More specifically, we select C from {1 10 3, 2 10 3, 5 10 3, 1 10 2, 2 10 2, 5 10 2 . . . , 1 103, 2 103}. For each C, we tune τ manually to make the rank of classifier matrix varied from 1 to the size of the matrix. We repeat this procedure ten times to compute the mean and standard deviation of the classification accuracy.