Kernel Principal Component Analysis (KPCA) is a widely-used feature extraction in pattern recognition. Considering that the conventional KPCA should decompose a kernel matrix of all training data, this would be an unrealistic assumption for data streams in real-world applications. Takeuchi et al. proposed an Incremental KPCA (IKPCA) where the learning is sequentially conducted for stream data. However, the eigenvalue decomposition should be carried out for every training data. In this paper, an extended IKPCA called Chunk IKPCA (CIKPCA) is proposed where a chunk of multiple data is learned at one time. Considering that not all data are useful for the eigen-feature space learning, the data in a chunk is first selected based on the accumulation ratio. In the proposed method, linearly independent data are selected from the reduced data and the eigen-feature space is spanned by the linear combination of these linearly independent data. Then, the eigenvalue decomposition is carried out only once for a chunk of selected data. In the experiments, we demonstrate the effectiveness of the proposed method.
展开▼