Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PLOOD: Partial Label Learning with Out-of-distribution Objects

Jintao Huang, Yiu-Ming Cheung, Chi-Man Vong

TL;DR

The paper addresses the fragility of traditional partial label learning (PLL) when Out-of-Distribution (OOD) samples appear, introducing the OODPLL paradigm and the PLOOD framework. PLOOD combines a Positive-Negative Sample Augmentation (PNSA) module for adaptive, perturbation-aware feature learning with a confidence-calibration mechanism and a Partial Energy (PE) score for integrated OOD detection and label refinement. Ablation studies confirm that both PNSA and PE are essential for robust performance, and experiments on CIFAR-10/100 and real-world PLL datasets show that PLOOD outperforms state-of-the-art PLL methods in open-set scenarios while maintaining strong in-distribution accuracy. The approach offers a practical, interpretable solution for open-set PLL, with broad implications for reliable weak supervision under distribution shifts.

Abstract

Existing Partial Label Learning (PLL) methods posit that training and test data adhere to the same distribution, a premise that frequently does not hold in practical application where Out-of-Distribution (OOD) objects are present. We introduce the OODPLL paradigm to tackle this significant yet underexplored issue. And our newly proposed PLOOD framework enables PLL to tackle OOD objects through Positive-Negative Sample Augmented (PNSA) feature learning and Partial Energy (PE)-based label refinement. The PNSA module enhances feature discrimination and OOD recognition by simulating in- and out-of-distribution instances, which employ structured positive and negative sample augmentation, in contrast to conventional PLL methods struggling to distinguish OOD samples. The PE scoring mechanism combines label confidence with energy-based uncertainty estimation, thereby reducing the impact of imprecise supervision and effectively achieving label disambiguation. Experimental results on CIFAR-10 and CIFAR-100, alongside various OOD datasets, demonstrate that conventional PLL methods exhibit substantial degradation in OOD scenarios, underscoring the necessity of incorporating OOD considerations in PLL approaches. Ablation studies show that PNSA feature learning and PE-based label refinement are necessary for PLOOD to work, offering a robust solution for open-set PLL problems.

PLOOD: Partial Label Learning with Out-of-distribution Objects

TL;DR

The paper addresses the fragility of traditional partial label learning (PLL) when Out-of-Distribution (OOD) samples appear, introducing the OODPLL paradigm and the PLOOD framework. PLOOD combines a Positive-Negative Sample Augmentation (PNSA) module for adaptive, perturbation-aware feature learning with a confidence-calibration mechanism and a Partial Energy (PE) score for integrated OOD detection and label refinement. Ablation studies confirm that both PNSA and PE are essential for robust performance, and experiments on CIFAR-10/100 and real-world PLL datasets show that PLOOD outperforms state-of-the-art PLL methods in open-set scenarios while maintaining strong in-distribution accuracy. The approach offers a practical, interpretable solution for open-set PLL, with broad implications for reliable weak supervision under distribution shifts.

Abstract

Existing Partial Label Learning (PLL) methods posit that training and test data adhere to the same distribution, a premise that frequently does not hold in practical application where Out-of-Distribution (OOD) objects are present. We introduce the OODPLL paradigm to tackle this significant yet underexplored issue. And our newly proposed PLOOD framework enables PLL to tackle OOD objects through Positive-Negative Sample Augmented (PNSA) feature learning and Partial Energy (PE)-based label refinement. The PNSA module enhances feature discrimination and OOD recognition by simulating in- and out-of-distribution instances, which employ structured positive and negative sample augmentation, in contrast to conventional PLL methods struggling to distinguish OOD samples. The PE scoring mechanism combines label confidence with energy-based uncertainty estimation, thereby reducing the impact of imprecise supervision and effectively achieving label disambiguation. Experimental results on CIFAR-10 and CIFAR-100, alongside various OOD datasets, demonstrate that conventional PLL methods exhibit substantial degradation in OOD scenarios, underscoring the necessity of incorporating OOD considerations in PLL approaches. Ablation studies show that PNSA feature learning and PE-based label refinement are necessary for PLOOD to work, offering a robust solution for open-set PLL problems.
Paper Structure (9 sections, 13 equations, 4 figures, 4 tables, 1 algorithm)

This paper contains 9 sections, 13 equations, 4 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Proposed Methodology
  3. PNSA Module: Adaptive Perturbation-Aware Feature Learning for OOD Differentiation
  4. Label Disambiguation via Confidence Calibration using PNSA Module
  5. Partial-Energy Score-based OOD Detection Integrated with Label Confidence
  6. Experiments
  7. Experimental Setup
  8. Experimental Results
  9. Conclusion

Figures (4)

  • Figure 1: Practical example of SOTA PLL methods on CIFAR-10 under No- and With- OOD obejcts.
  • Figure 2: An illustration of OODPLL. Top-left is a scenario for training in animal image classification. Each image possesses a collection of candidate labels, including Cat, Fox, and Dog; however, only Cat represents the accurate label. Bottom-left are the images of in-distribution (ID) and out-of-distribution (OOD) animals. In these instances, the OODPLL task must identify the true labels of the animals (ID objects) and detect the non-animal category (i.e., OOD objects).
  • Figure 3: Accuracy of six SOTA PLL methods on five practical PLL datasets with OOD objects.
  • Figure 4: Convergence Analysis of PLOOD On CIFAR-100.