Abstract:The proliferation and ubiquity of temporal data across many disciplines has sparked interest for similarity, classification and clustering methods specifically designed to handle time series data. A core issue when dealing with time series is determining their pairwise similarity, i.e., the degree to which a given time series resembles another. Traditional distance measures such as the Euclidean are not well-suited due to the time-dependent nature of the data. Elastic metrics such as dynamic time warping (DTW) offer a promising approach, but are limited by their computational complexity, non-differentiability and sensitivity to noise and outliers. This thesis proposes novel elastic alignment methods that use parametric \& diffeomorphic warping transformations as a means of overcoming the shortcomings of DTW-based metrics. The proposed method is differentiable \& invertible, well-suited for deep learning architectures, robust to noise and outliers, computationally efficient, and is expressive and flexible enough to capture complex patterns. Furthermore, a closed-form solution was developed for the gradient of these diffeomorphic transformations, which allows an efficient search in the parameter space, leading to better solutions at convergence. Leveraging the benefits of these closed-form diffeomorphic transformations, this thesis proposes a suite of advancements that include: (a) an enhanced temporal transformer network for time series alignment and averaging, (b) a deep-learning based time series classification model to simultaneously align and classify signals with high accuracy, (c) an incremental time series clustering algorithm that is warping-invariant, scalable and can operate under limited computational and time resources, and finally, (d) a normalizing flow model that enhances the flexibility of affine transformations in coupling and autoregressive layers.

Residual Networks as Flows of Velocity Fields for Diffeomorphic Time Series Alignment

Diffeomorphic Transformations for Time Series Analysis: An Efficient Approach to Nonlinear Warping

DTW-NN: A novel neural network for time series recognition using dynamic alignment between inputs and weights

TLRN: Temporal Latent Residual Networks For Large Deformation Image Registration

Learning spatial-temporal deformable networks for unconstrained face alignment and tracking in videos

Deep Declarative Dynamic Time Warping for End-to-End Learning of Alignment Paths

Affine and Regional Dynamic Time Warpng

Flow-Based End-to-End Model for Hierarchical Time Series Forecasting via Trainable Attentive-Reconciliation

Modeling Spatial Nonstationarity via Deformable Convolutions for Deep Traffic Flow Prediction

DifFormer: Multi-Resolutional Differencing Transformer With Dynamic Ranging for Time Series Analysis

metricDTW: local distance metric learning in Dynamic Time Warping

Radflow: A Recurrent, Aggregated, and Decomposable Model for Networks of Time Series

Multiscale Manifold Warping

C2F-FWN: Coarse-to-Fine Flow Warping Network for Spatial-Temporal Consistent Motion Transfer

Transformer network with decoupled spatial–temporal embedding for traffic flow forecasting

TransFlowNet: A Physics-Constrained Transformer Framework for Spatio-Temporal Super-Resolution of Flow Simulations.

Convergence Analysis of Flow Matching in Latent Space with Transformers

End-to-End Modeling of Hierarchical Time Series Using Autoregressive Transformer and Conditional Normalizing Flow-based Reconciliation

Dynamic Time Warp Convolutional Networks

Spatio-Temporal Alignments: Optimal transport through space and time

Multilevel Wavelet Decomposition Network for Interpretable Time Series Analysis