ImputeINR: Enhancing Time Series Imputation with Adaptive Group-based Implicit Neural Representations

MENGXUAN LI, Ke Liu, Jialong Guo, Jiajun Bu, Hongwei Wang, Haishuai Wang

Keywords: time series imputation, implicit neural representations

TL;DR: We propose ImputeINR, a time series imputation method based on adaptive group-based implicit neural representations.

Abstract: Time series data frequently exhibit the presence of missing values, rendering imputation a crucial process for downstream time series tasks and applications. However, existing imputation methods focus on discrete data points and are unable to effectively model sparse data, resulting in particularly poor performance for imputing substantial missing values. In this paper, we propose a novel approach, ImputeINR, for time series imputation by employing implicit neural representations (INR) to learn continuous functions for time series. ImputeINR leverages the merits of INR that the continuous functions are not coupled to sampling frequency and have infinite sampling frequency, allowing ImputeINR to generate fine-grained imputations even on extremely absent observed values. In addition, we introduce a multi-scale feature extraction module in ImputeINR architecture to capture patterns from different time scales, thereby effectively enhancing the fine-grained and global consistency of the imputation. To address the unique challenges of complex temporal patterns and multiple variables in time series, we design a specific form of INR continuous function that contains three additional components to learn trend, seasonal, and residual information separately. Furthermore, we innovatively propose an adaptive group-based framework to model complex residual information, where variables with similar distributions are modeled by the same group of multilayer perception layers to extract necessary correlation features. Since the number of groups and their output variables are determined by variable clustering, ImputeINR has the capacity of adapting to diverse datasets. Extensive experiments conducted on seven datasets with five ratios of missing values demonstrate the superior performance of ImputeINR, especially for high absent ratios in time series.

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.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 9758