SHIFT-RESILIENT DIFFUSIVE IMPUTATION FOR VARIABLE SUBSET FORECASTING

Haihua XU, Qi Hao, He zhang, Jianpeng Zhao, Ziyue Qiao, Lu Jiang, pengfei wang, Pengyang Wang

Keywords: data mining; variable subset forecasting; distribution shift

Abstract: It is common for sensor failures to result in missing data, leading to training sets being complete while test sets have only a small subset of variables. The challenge lies in utilizing incomplete data for forecasting, which is known as the Variable Subset Forecasting (VSF). In VSF tasks, significant distribution shift is present. One type is inter-series shift, which indicates changes in correlations between different series, and the other type is intra-series shift, which refers to substantial distribution differences within the same series across different time windows. Existing approaches to solving VSF tasks typically involve imputing the missing data first and then making predictions using the completed series. However, these methods do not account for the shift inherent in VSF tasks, resulting in poor model performance. To address these challenges, we propose a Shift-Resilient Diffusive Imputation (SRDI) framework against the shift. Specifically, SRDI integrates divide-conquer strategy with the denoising process, that decomposes the input into invariant patterns and variant patterns, representing the temporally stable parts of inter-series correlation and the highly fluctuating parts, respectively. By extracting spatiotemporal features from each separately and then appropriately combining them, inter-series shift can be effectively mitigated. Then, we innovatively organize SRDI and the forecasting model into a meta-learning paradigm tailored for VSF scenarios. We address the intra-series shift by treating time windows as tasks during training and employing an adaptation process before testing. Extensive experiments on four datasets have demonstrated our superior performance compared with state-of-the-art methods.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3462