Abstract:Modeling stochastic and irregularly sampled time series is a challenging problem found in a wide range of applications, especially in medicine. Neural stochastic differential equations (Neural SDEs) are an attractive modeling technique for this problem, which parameterize the drift and diffusion terms of an SDE with neural networks. However, current algorithms for training Neural SDEs require backpropagation through the SDE dynamics, greatly limiting their scalability and stability. To address this, we propose Trajectory Flow Matching (TFM), which trains a Neural SDE in a simulation-free manner, bypassing backpropagation through the dynamics. TFM leverages the flow matching technique from generative modeling to model time series. In this work we first establish necessary conditions for TFM to learn time series data. Next, we present a reparameterization trick which improves training stability. Finally, we adapt TFM to the clinical time series setting, demonstrating improved performance on three clinical time series datasets both in terms of absolute performance and uncertainty prediction.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the challenge of modeling randomly and irregularly sampled time series in medical and other fields. Specifically, the authors propose the **Trajectory Flow Matching (TFM)** method to overcome several key problems in existing neural stochastic differential equations (Neural SDEs) training methods: 1. **Limitations of existing methods**: - Current algorithms for training Neural SDEs need to back - propagate through SDE dynamics, which greatly limits their scalability and stability. - These methods are difficult to handle cross - trajectories and complex covariate effects, especially in medical time - series data. 2. **The proposed new method**: - **Trajectory Flow Matching (TFM)**: This method can train Neural SDEs without simulation, bypassing the need for back - propagation through dynamics. It uses the flow - matching technique in generative models to model time series. - TFM improves training stability by introducing the conditional re - parameterization trick and adapts to the clinical time - series setting, showing improved performance on three clinical time - series datasets, especially in terms of absolute performance and uncertainty prediction. 3. **Specific application scenarios**: - The paper specifically focuses on time - series modeling in the medical field, such as the evolution path of patients' vital signs. These paths may be significantly different or even cross due to the influence of covariates such as drug interventions and underlying health conditions. Therefore, accurate modeling of these complex time series is crucial for dynamic monitoring and care. 4. **Main contributions**: - It is proved that under certain conditions, continuous - time dynamics can be trained without simulation using the matching technique. - The method is extended to handle irregularly sampled trajectories, and uncertainty is estimated using the uncertainty prediction loss. - Empirical studies show that when this method is applied to actual clinical data, the error is reduced by 15% to 83%. Through these improvements, TFM provides a more efficient and stable method for handling complex time - series data, especially those with high uncertainty and irregular sampling characteristics.

Trajectory Flow Matching with Applications to Clinical Time Series Modeling

FM-TS: Flow Matching for Time Series Generation

Time-changed normalizing flows for accurate SDE modeling

Meta Flow Matching: Integrating Vector Fields on the Wasserstein Manifold

Efficient Trajectory Forecasting and Generation with Conditional Flow Matching

Flow Matching with Gaussian Process Priors for Probabilistic Time Series Forecasting

Conditional Variable Flow Matching: Transforming Conditional Densities with Amortized Conditional Optimal Transport

ImageFlowNet: Forecasting Multiscale Image-Level Trajectories of Disease Progression with Irregularly-Sampled Longitudinal Medical Images

Stream-level flow matching from a Bayesian decision theoretic perspective

Trajectory Forecasting with Neural Networks: an Empirical Evaluation and A New Hybrid Model

Flow Matching for Scalable Simulation-Based Inference

Modeling single cell trajectory using forward-backward stochastic differential equations

Stochastic Flow Matching for Resolving Small-Scale Physics

Multivariate Probabilistic Time Series Forecasting via Conditioned Normalizing Flows

Manifold Interpolating Optimal-Transport Flows for Trajectory Inference

Flow-based Spatio-Temporal Structured Prediction of Motion Dynamics

Probabilistic Temporal Prediction of Continuous Disease Trajectories and Treatment Effects Using Neural SDEs

Neural network stochastic differential equation models with applications to financial data forecasting

TrajGPT: Irregular Time-Series Representation Learning for Health Trajectory Analysis

Unifying Bayesian Flow Networks and Diffusion Models through Stochastic Differential Equations