| 摘要 |
PROBLEM TO BE SOLVED: To accurately detect a change point in multi-dimensional time series signals from signals at a small number of samples.SOLUTION: An extraction and provision unit 12 extracts a signal X(t) before the time point t and a signal X(t+n) after the time point t from multi-dimensional time series signals including a plurality of signals, and provides a label of y=+1 to X(t) and label of y=-1 to X(t+n). An importance calculation unit 14 calculates, as the importance of the signals with respect to a change point, a variable α, which is the solution of an optimization problem including a regularization term using X(t), X(t+n), y, and the L1 norm of a variable α. A change point determination unit 16 determines whether or not a change point is present in the multi-dimensional time series signals at a predetermined time point on the basis of a score using the importance αand a value indicating the independence obtained by an HSIC formed by the signals and y. |
