UAKNN: Label Distribution Learning via Uncertainty-Aware KNN
Pu Wang, Yu Zhang, Zhuoran Zheng
TL;DR
This work tackles Label Distribution Learning by proposing UAKNN, a parameter-free, uncertainty-aware KNN that exploits a low-rank prototype space and micro-perturbation to robustly estimate label distributions without heavy training. The method computes uncertainty-based weights for prototype-driven samples and fuses them through a Softmax^{*}-based normalization to yield calibrated distributions, enabling real-time GPU-accelerated inference. Extensive experiments across 12 benchmarks show competitive accuracy and speed, with ablations confirming the value of prototype matching and perturbation, and specialized strategies for extreme-label cases like Gene. The approach offers practical online deployability and incremental learning benefits, reducing training costs while maintaining strong LDL performance.
Abstract
Label Distribution Learning (LDL) aims to characterize the polysemy of an instance by building a set of descriptive degrees corresponding to the instance. In recent years, researchers seek to model to obtain an accurate label distribution by using low-rank, label relations, expert experiences, and label uncertainty estimation. In general, these methods are based on algorithms with parameter learning in a linear (including kernel functions) or deep learning framework. However, these methods are difficult to deploy and update online due to high training costs, limited scalability, and outlier sensitivity. To address this problem, we design a novel LDL method called UAKNN, which has the advantages of the KNN algorithm with the benefits of uncertainty modeling. In addition, we provide solutions to the dilemma of existing work on extremely label distribution spaces. Extensive experiments demonstrate that our method is significantly competitive on 12 benchmarks and that the inference speed of the model is well-suited for industrial-level applications.