Abstract
Temporal alignment of sequences is a fundamental challenge in manyapplications, such as computer vision and bioinformatics, where local timeshifting needs to be accounted for. Misalignment can lead to poor modelgeneralization, especially in high-dimensional sequences. Existing methodsoften struggle with optimization when dealing with high-dimensional sparsedata, falling into poor alignments. Feature selection is frequently used toenhance model performance for sparse data. However, a fixed set of selectedfeatures would not generally work for dynamically changing sequences and wouldneed to be modified based on the state of the sequence. Therefore, modifyingthe selected feature based on contextual input would result in betteralignment. Our suggested method, Conditional Deep Canonical Temporal TimeWarping (CDCTW), is designed for temporal alignment in sparse temporal data toaddress these challenges. CDCTW enhances alignment accuracy for highdimensional time-dependent views be performing dynamic time warping on dataembedded in maximally correlated subspace which handles sparsity with novelfeature selection method. We validate the effectiveness of CDCTW throughextensive experiments on various datasets, demonstrating superior performanceover previous techniques.