http://ieeexplore.ieee.org TOC Alert for Publication# 6046 2012 February 10 14 1 C1 C4 49 14 1 C2 C2 35 14 1 1 2 34 p and small sample size n (large p small n ) between pairs of video-audio modalities. We demonstrate the superiority of the SODA approach in video indexing, retrieval, and activity recognition as compared to the state-of-the-art methods such as hidden Markov models (HMM), support vector machine (SVM), cross-media indexing space (CMIS), and other noncausal divergence measures such as mutual information (MI). We also demonstrate the success of SODA in audio and video localization and indexing/retrieval of data with missaligned modalities.]]> 14 1 3 16 1116 14 1 17 27 1077 14 1 28 42 2390 14 1 43 54 1642 l₁ regularized least squares). We apply the parallel incomplete Cholesky factorization to approximate the covariance statistics in the preprocess step, and the parallel primal-dual interior-point method with the Sherman-Morrison-Woodbury formula to optimize the model parameters. For a dataset with n samples in a d-dimensional space, compared with lasso, Plasso significantly reduces complexities from the original O(d³) for computational time and O(d²) for storage space to O(h²d/m) and O(hd/m) , respectively, if the system has m processors and the reduced dimension h is much smaller than the original dimension d . Furthermore, we develop the kernel extension of the proposed linear algorithm with the sample reweighting schema, and we can achieve similar time and space complexity improvements [time complexity from O(n³) to O(h²n/m) and the space complexity from O(n²) to O(hn/m), for a dataset with n training examples]. Experimental results on TRECVID video concept detection challenges suggest that the proposed method can obtain significant time and space savings for training effective detectors with limited communication overhead.]]> 14 1 55 65 1192 O(1/K²) (where K is the iteration counter), in contrast to traditional sub-gradient descent methods of O(1/√K). Our method provides a scalable solution for both key frame extraction and video skim generation, because one can select an arbitrary number of key frames to represent the original videos. Experiments on a human labeled benchmark dataset and comparisons to the state-of-the-art methods demonstrate the advantages of our algorithm.]]> 14 1 66 75 2141 14 1 76 87 981 14 1 88 101 2204 14 1 102 110 745 14 1 111 120 922 14 1 121 128 1836 14 1 129 141 1255 14 1 142 156 1537 14 1 157 167 1199 14 1 168 177 1101 14 1 178 186 911 14 1 187 198 1579 14 1 199 210 2388 14 1 211 222 1165 14 1 223 233 910 14 1 234 234 16 14 1 235 236 46 14 1 C3 C3 28