Hopformer: Homogeneity-Pursuit Transformer for Time Series Forecasting
Abstract
Forecasting multiple time-series with high-dimensional covariates presents a core challenge: unifying common temporal patterns while retaining meaningful series-specific information. We introduce Hopformer (Homogeneity-Pursuit Transformer), a two-stage framework that addresses this challenge. In the first stage, we perform a Sparsity Pattern Aggregation (SPA) scheme extracting a common low-variance trend that incorporates the covariates. This acts as a homogenization layer. In the second stage, a LoRA-fine-tuned Transformer models the remaining complex dependencies in the residual. Our method is theoretically grounded. We prove that SPA achieves a near-optimal bias-variance trade-off via an oracle inequality. We also provide generalization bounds for the second stage under dependent time series data. Hopformer sets a new state of the art, improving MASE by an average of 6.56% across synthetic and real-world forecasting benchmarks.
Related
- Homogeneity Pursuit for 3D Point Clouds
- Homogeneity Pursuit of Panel Data with Sparsity
- FHRFormer: A Self-Supervised Masked Transformer Framework for Fetal Heart Rate Time-Series Inpainting and Forecasting
- RATSIFT: Resolution-Adaptive Transformer for Solar Irradiance Forecasting and Temporal Downscaling
- STCFormer: Adaptive Spatio-Temporal Modeling with Dynamic Cluster Transformer for Station-based Weather Forecasting
BibTeX
@misc{zhang2026hopformer,
title = {{Hopformer: Homogeneity-Pursuit Transformer for Time Series Forecasting}},
author = {Wan Zhang and Qinjie Lin and Chan Lee and Weijian Li and Han Liu and Kai Zhang},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.22299},
eprint = {2607.22299},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.22299},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Hopformer: Homogeneity-Pursuit Transformer for Time Series Forecasting
AU - Wan Zhang
AU - Qinjie Lin
AU - Chan Lee
AU - Weijian Li
AU - Han Liu
AU - Kai Zhang
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2607.22299
UR - https://arxiv.org/abs/2607.22299
AB - Forecasting multiple time-series with high-dimensional covariates presents a core challenge: unifying common temporal patterns while retaining meaningful series-specific information. We introduce Hopformer (Homogeneity-Pursuit Transformer), a two-stage framework that addresses this challenge. In the first stage, we perform a Sparsity Pattern Aggregation (SPA) scheme extracting a common low-variance trend that incorporates the covariates. This acts as a homogenization layer. In the second stage, a LoRA-fine-tuned Transformer models the remaining complex dependencies in the residual. Our method is theoretically grounded. We prove that SPA achieves a near-optimal bias-variance trade-off via an oracle inequality. We also provide generalization bounds for the second stage under dependent time series data. Hopformer sets a new state of the art, improving MASE by an average of 6.56% across synthetic and real-world forecasting benchmarks.
ER -