Dynamic Time Warping

Dynamic Time Warping (DTW) finds a low-cost monotone alignment path between sequences, allowing similar events to occur at different speeds or times. Researchers use it for time-series classification and averaging, speech and music comparison, gait analysis, trajectory transfer, and comparisons of temporal graphs; common evaluations include UCR datasets and application-specific recordings or simulations. Approaches range from discrete DTW and context-aware variants to differentiable soft relaxations, continuous or diffeomorphic warps, and GPU-accelerated implementations. A central challenge is balancing flexible, robust alignment and computational efficiency with preserving meaningful temporal structure, such as transitions, physical relationships, or local timing cues.