Probabilistic Feature Smoothed Gaussian Process For Imbalanced Regression

Yifei Wu, Heyue Li, Xinbing Wang, Qinying Gu, Nanyang Ye

Keywords: Imbalanced Learning, Gaussian Process, Bayesian Methods, Machine Learning

Abstract: Gaussian Processes (GPs) are non-parametric Bayesian models widely used for regression, classification, and other tasks due to their explainability and versatility. However, GPs face challenges in imbalanced regression, where the skewed distribution of target labels can greatly harm models' performances. In this work, we introduce the Probabilistic Feature Smoothed Partially Independent Training Conditional Approximation (PFS-PITC) to enhance GP performance in imbalanced scenarios. We extract statistical features from the observation space using equidistant label intervals and apply kernel smoothing to address sampling density discontinuities. This process enables PFS-PITC to utilize information from nearby labels within imbalanced datasets, thereby reducing GPs' sensitivity to such imbalances. Empirical tests on various imbalanced regression datasets demonstrate the effectiveness of PFS-PITC, contributing to the robustness of GPs in handling flawed real-world data and expanding their applicability in challenging data processing tasks.

Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)

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: 2588