Improved Learning via k-DTW: A Novel Dissimilarity Measure for Curves
Improved Learning via k-DTW: A Novel Dissimilarity Measure for Curves
This paper introduces $k$-Dynamic Time Warping ($k$-DTW), a novel dissimilarity measure for polygonal curves. $k$-DTW has stronger metric properties than Dynamic Time Warping (DTW) and is more robust to outliers than the Fr\'{e}chet distance, which are the two gold standards of dissimilarity measures for polygonal curves. We show interesting properties of $k$-DTW and give an exact algorithm as well as a $(1+\varepsilon)$-approximation algorithm for $k$-DTW by a parametric search for the $k$-th largest matched distance. We prove the first dimension-free learning bounds for curves and further learning theoretic results. $k$-DTW not only admits smaller sample size than DTW for the problem of learning the median of curves, where some factors depending on the curves' complexity $m$ are replaced by $k$, but we also show a surprising separation on the associated Rademacher and Gaussian complexities: $k$-DTW admits strictly smaller bounds than DTW, by a factor $\tilde\Omega(\sqrt{m})$ when $k\ll m$. We complement our theoretical findings with an experimental illustration of the benefits of using $k$-DTW for clustering and nearest neighbor classification.
Amer Krivo?ija、Alexander Munteanu、André Nusser、Chris Schwiegelshohn
计算技术、计算机技术
Amer Krivo?ija,Alexander Munteanu,André Nusser,Chris Schwiegelshohn.Improved Learning via k-DTW: A Novel Dissimilarity Measure for Curves[EB/OL].(2025-05-29)[2025-06-18].https://arxiv.org/abs/2505.23431.点此复制
评论