iQR: Quantile Regression with QR Orthogonal Decomposition for Resource Scheduling Optimization without Empirical Model

Yanyan He, Xiaoyi Liu, Duxin Chen, Wenwu Yu

Keywords: Resource Scheduling Strategy, Sparse System Identification, L1-norm Quantile Regression, Business Constraint Satisfaction

TL;DR: This work presents a resource scheduling strategy based on quantile regression and QR orthogonal decomposition designed to meet the demand at the lowest cost, with results on multiple datasets showing its superiority over neural network approaches.

Abstract: Optimal resource scheduling aims to cover resource demand with minimum economic cost, which is far from model-based constraint optimization or data-driven prediction based scheduling process. To address this model-free constraint optimization issue, a sparse system identification framework with quantile regression and QR orthogonal decomposition~(iQR) is proposed for complex systems ranging from small to large scales. It leverages quantile optimization with $L_1$-norm to construct a business-driven strategy to reach the proportion of meeting resource demand. It also involves a complete-mapping Fourier Transformation process and an orthogonal least squares technique to select basis vectors in advance to achieve fast regression with sparse mathematical expression, which reduces the number of basis functions from thousands to hundreds and to dozens. iQR represents a specific expectation for single time series prediction, which only achieves predictions that deviate from the true values as little as possible, but aims for predictions consistently higher than the actual values of real demand. Numerical experiments was conducted on eight datasets, including commonly used time series and real-world CPU resource data. The results indicate that most neural network-based methods fail to balance both resource demands and prediction accuracy effectively. In contrast, iQR can achieve optimal scheduling with the minimum economic cost and it is easier to satisfy the business constraints with quantile tuning. Notably, iQR is lightweight with a training speed in seconds and does not rely on the support of computing power of GPU resources. This study may provide new insight into investigations on resource scheduling optimization issues.

Primary Area: other topics in machine learning (i.e., none of the above)

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: 1697