arXiv Machine Learning

Aurora-X: Built for Extreme Time Series Forecasting

Aurora‑X is a billion‑parameter time‑series foundation model designed for extreme forecasting tasks. It employs a progressive curriculum that starts with channel‑independent pretraining, then adds cross‑variable dependencies, variable context and horizon lengths, and optional future covariates during mid‑training. A variable‑resolution post‑training stage allows adjustable temporal spans per token at inference, while a pattern‑guided mixture‑of‑experts expands capacity through sparse activation and expert specialization. An implicit quantile network head predicts arbitrary quantiles, enhancing probabilistic forecasting flexibility. Experiments on GIFT‑Eval, TIME, FEV‑Bench, TFB, and DAG‑Bench show state‑of‑the‑art performance against both pretrained TSFMs and task‑specific supervised models.

arXiv AI
Oct 9

Timer-M1: A Multivariate Time Series Foundation Model via Learning Primitives

Timer‑M1 is a pretrained multivariate time‑series foundation model that learns with primitives for zero‑shot forecasting. It uses a primitive‑based data synthesis and pretraining pipeline to generate series with shared temporal and relational patterns, assembling them into multivariate samples with distinct channel roles. The model employs gated two‑dimensional Transformer blocks that dynamically allocate cross‑variate attention, achieving top performance on several large‑scale forecasting benchmarks.

By Haoran Zhang, Haixuan Liu, Xingjian Su, Yong Liu, Zhi Chen, Yuxuan Wang, Jianmin Wang, Mingsheng Long
arXiv Machine Learning
Oct 28

GateTS: Versatile and Efficient Forecasting via Attention-Inspired routed Mixture-of-Experts

GateTS introduces a streamlined univariate forecasting model that merges sparse Mixture-of-Experts computation with an attention-inspired gating mechanism, eliminating the need for auxiliary load‑balancing losses. The architecture achieves superior predictive accuracy across both short‑ and long‑term horizons, including intermittent signals, while using far fewer parameters than leading transformer models such as PatchTST. Empirical tests on diverse datasets show GateTS is computationally more efficient than LSTM, enabling cost‑effective inference for real‑world time‑series applications.

By Kyrylo Yemets, Mykola Lukashchuk, Ivan Izonin
arXiv AI
1d ago

FlexCast: Flexible Context Learning for Time Series Foundation Models

FlexCast introduces a flexible context-target learning framework for time series foundation models, reformulating forecasting as variable-length masked prediction over time series patches. By dynamically sampling visible context and masked target patches during pretraining, it exposes the model to diverse context-target configurations without sequential decoding. The same pretrained model can perform one-step forecasting, confidence-based iterative refinement, and next-token-style generation, achieving competitive or superior accuracy to strong TSFM baselines while reducing inference latency.

By Zuyang Shen, Luqi Gong, Fanda Fan, Bingsong Wang, Xiang Ao, Yahui Wei, Minghao Liu, Chunjie Luo, Jianfeng Zhan
arXiv Machine Learning
Sep 24

A Foundation Model for Instruction-Conditioned In-Context Time Series Tasks

The paper introduces iAmTime, a time‑series foundation model that uses instruction‑conditioned in‑context learning to adapt to tasks at inference time. iAmTime represents each episode as a structured prompt with semantic tokens that focus on specific time‑series regions, enabling the model to infer task structure from input‑output demonstrations. Trained on large real and synthetic corpora across forecasting, imputation, reconstruction, classification, anomaly detection, and source de‑mixing, iAmTime outperforms strong baselines on zero‑shot probabilistic and point forecasting while matching or exceeding performance on several non‑forecasting tasks.

By Anish Saha, Konstantin Shmakov
arXiv Machine Learning
Sep 15

Tabby: An Open Pretraining Recipe for Time Series Foundation Models

arXiv:2609.13956v1 Announce Type: new Abstract: In this report, we release Tabby, a long context probabilistic time series foundation model, together with a complete and open recipe of how it was bui...

By Shifeng Xie, Bahaeddine Abdessalem, Zehao Xiao, Youssef Attia El Hili, Ambroise Odonnat, Zhiwei Dong, Lei Zan, Themis Palpanas, Jianfeng Zhang, Lujia Pan, Keli Zhang, Malik Tiomoko
arXiv Machine Learning
1d ago

Time Flies When You Parallelise: Pareto-efficient Time Series Foundation Models with Linear Recurrences

The paper introduces ci‑proto, a family of time‑series foundation models built on Mamba‑3 that use linear state recurrences to enable chunk‑wise parallel‑in‑time computation. This design reduces training and inference costs to linear scaling with sequence length while alternating recurrent temporal processing with cross‑variate attention for multivariate probabilistic forecasting. On fev‑bench’s 100 forecasting tasks, ci‑proto XL achieves state‑of‑the‑art performance, and the five ci‑proto variants collectively form the Pareto frontier for mean scaled quantile loss and mean absolute scaled error versus parameter count and median runtime.

By Torben Berndt, Lingyi Yang, Federico Nardi, Yi Heng Lim, AJ Shears, Benjamin Walker
arXiv AI
Oct 2

ProtoFlow: Prototype-Guided Flow Matching for Multivariate Time Series Forecasting

ProtoFlow is a new multivariate time series forecasting framework that combines vector‑quantized autoencoding with prototype‑guided flow matching. It maps sequences into a discrete latent space, constructs a structured prior from the learned VQ codebook, and trains a DiT‑based rectified flow to transport samples from this prior to future latent representations conditioned on past observations. By replacing generic Gaussian noise with a learned prototype prior, ProtoFlow eliminates autoregressive rollout mismatch and achieves faster training convergence while delivering superior forecasting performance on benchmark datasets.

By Shibo Feng, Wanjin Feng, Yang Qiu, Deheng Ye, Peilin Zhao, Chunyan Miao