Search papers, labs, and topics across Lattice.
This paper introduces WrapFlow, a novel continuous-time modeling framework designed to enhance irregular time series forecasting by addressing the challenges of asynchronous observations and non-uniform sampling intervals. By employing Continuous-Time Tokenization to encode raw observation events and gap-aware tokens, WrapFlow effectively captures long-range temporal dependencies using a standard Transformer architecture. The framework's simulation-free training paradigm for Residual Flow Matching allows for high-quality forecasting with minimal computational overhead, achieving state-of-the-art results across various real-world datasets.
WrapFlow revolutionizes irregular time series forecasting by eliminating the need for complex numerical solvers while achieving state-of-the-art accuracy.
Irregular multivariate time series are widely encountered in applications such as healthcare monitoring, human activity recognition, and environmental sensing. Their core challenges stem from asynchronous observations, non-uniform sampling intervals, and the fact that temporal patterns themselves carry critical dynamic information. Existing approaches either rely on discretization-based preprocessing (e.g., interpolation, imputation, or aggregation), which disrupts the underlying continuous-time semantics, or adopt continuous-time modeling via ODE-based frameworks, which typically require specialized architectures and incur substantial computational overhead due to numerical solvers. To address these limitations, we propose WrapFlow, a continuous-time modeling framework for irregular time series forecasting. On the input side, WrapFlow introduces Continuous-Time Tokenization, which directly encodes raw observation events and explicitly models long unobserved intervals via gap-aware tokens. The resulting continuous-time tokens are then processed by a standard Transformer backbone to capture long-range temporal dependencies. On the output side, we develop a simulation-free training paradigm for Residual Flow Matching, which learns conditional residual vector fields around base predictions while avoiding numerical-solver simulation and backpropagation during training. This design enables high-quality continuous forecasting using only a small number of fixed rollout steps at inference. Extensive experiments on multiple real-world datasets demonstrate that WrapFlow achieves state-of-the-art performance.
