Lightweight time series forecasting reduces latency and saves computation
CTLite: Lightweight Irregular Multivariate Time Series Forecasting with Reusable Continuous-Time Summaries
Machine Learning
Summary
Irregular time series data, where measurements happen at uneven times and from multiple sources, is hard to predict. The authors designed CTLite, a new way to make predictions that separates summarizing past data from learning to predict, which is more efficient. Their method creates reusable summaries of the past that don’t depend on training steps, so repeated predictions are faster without losing accuracy. Tests on real-world data show CTLite matches or beats other methods with fewer parameters and much less computing time.
What this means in practice
- •For industrial iot engineers: Generate real-time forecasts from sparse sensor streams with less computing power and lower prediction delay.
- •For health monitoring teams: Make faster, accurate predictions from irregular patient data collected at varying times and sources.
Authors
Shusen Ma, Xudong Zuo, Yangchun Xie, Yun-Bo Zhao
Abstract
Irregular multivariate time series forecasting (IMTSF) requires capturing temporal dynamics and cross-variable dependencies from sparse, asynchronous observations. Existing methods often rely on learned temporal encoders, coupling historical representation construction with model optimisation and repeatedly processing the same observations. We propose CTLite, a lightweight forecasting framework that decouples historical summarisation from learned prediction. Specifically, a parameter-free continuous-time moment encoder summarises local levels and trends at multiple temporal scales while suppressing poorly supported slope estimates. A compact readout combines damped local extrapolation with a low-rank cross-variable correction that accounts for observation freshness, target availability, and forecast horizon. Since the historical summaries are independent of trainable parameters and forecast queries, they can be reused across optimisation steps and repeated predictions for an unchanged history without altering predictions or parameter gradients. Experiments on four real-world benchmarks show that CTLite achieves the best or second-best mean MSE and MAE among the evaluated methods with only 88-2,537 trainable parameters. Under matched GPU workloads at batch size 32, CTLite reduces inference latency by 18.6-87.3% compared with the latest models, including the cost of historical encoding. Reusing summaries further reduces total runtime by approximately 52% over 16 forecast calls on the same history compared with recomputing the encoder for every call, including the initial encoding cost. The code is at: https://github.com/SSMa913/NICLab-CTLite.