aiqbench analysis
SurF introduces a novel generative framework that leverages the Time Rescaling Theorem to unify modeling of heterogeneous, irregularly sampled event streams like earthquakes and social media activity. By transforming variable inter-event intervals into unit-rate noise via a learnable bijection, this approach overcomes limitations in traditional tokenization methods while scaling effectively through Transformer-based pretraining. This work represents a significant advancement in handling asynchronous data for foundation models, offering improved forecasting accuracy across diverse real-world benchmarks.
SurF: A Generative Model for Multivariate Irregular Time Series Forecasting
arXiv:2605.14069v2 Announce Type: replace Abstract: Irregularly sampled multivariate event streams remain a difficult modality for generative modeling: tokenization-based approaches break down when inter-event intervals vary by orders of magnitude. We (i) propose \textbf{SurF}, a generative model that uses the Time Rescaling Theorem (TRT) as a learnable bijection between event sequences and i.i.d.\ unit-rate exponential noise, enabling a single model to be trained across heterogeneous event-stream datasets; (ii) three efficient parameterizations of the cumulative intensity that scale to long sequences; and (iii) a Transformer-based encoder for multi-dataset pretraining. On six real-world benchmarks, SurF achieves the best reported time RMSE on Earthquake, Retweet, and Taobao, and is within trial-level noise of the strongest specialist on the remaining three. Under a strict leave-one-out protocol, the held-out checkpoint beats every classical and neural-autoregressive baseline on $5/6$ datasets and beats every baseline on Amazon and Earthquake, an initial step toward foundation models over asynchronous event streams (Code is available at https://github.com/MrRezaeiUofT/SurF).
Abstract:Irregularly sampled multivariate event streams remain a difficult modality for generative modeling: tokenization-based approaches break down when inter-event intervals vary by orders of magnitude. We (i) propose \textbf{SurF}, a generative model that uses the Time Rescaling Theorem (TRT) as a learnable bijection between event sequences and i.i.d.\ unit-rate exponential noise, enabling a single model to be trained across heterogeneous event-stream datasets; (ii) three efficient parameterizations of the cumulative intensity that scale to long sequences; and (iii) a Transformer-based encoder for multi-dataset pretraining. On six real-world benchmarks, SurF achieves the best reported time RMSE on Earthquake, Retweet, and Taobao, and is within trial-level noise of the strongest specialist on the remaining three. Under a strict leave-one-out protocol, the held-out checkpoint beats every classical and neural-autoregressive baseline on $5/6$ datasets and beats every baseline on Amazon and Earthquake, an initial step toward foundation models over asynchronous event streams (Code is available at this https URL).
Machine Learning (cs.LG)
arXiv:2605.14069 [cs.LG]
(or arXiv:2605.14069v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.14069 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 245 words
Related coverage
Latent Laplace Diffusion for Irregular Multivariate Time Series
arXiv · Jun 3, 2026
QuITE: Query-Based Irregular Time Series Embedding
arXiv · Jun 3, 2026
FADTI: Fourier and Attention Driven Diffusion for Multivariate Time Series Imputation
arXiv · Jun 9, 2026
7 Steps to Mastering Time Series Analysis with Python
KDnuggets · Jun 4, 2026