PSDC: A Prototype-Based Shared-Dummy Classifier Model for Open-Set Domain Adaptation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Open-set domain adaptation (OSDA) addresses knowledge transfer challenges with domain and label shifts, including unknown target classes.
• Existing OSDA methods often treat all unknown instances as a single group, which is insufficient for distinct categories and distributions.
• Current approaches may overlook categorical discriminative information, hindering the learning of clear class boundaries in the target domain.

Purpose of the Study:

• To propose a novel prototype-based shared-dummy classifier (PSDC) model for addressing limitations in open-set domain adaptation.
• To improve the handling of unknown target classes and enhance discriminative boundary learning in OSDA.
• To achieve state-of-the-art performance in OSDA tasks by overcoming existing method drawbacks.

Main Methods:

• Introduced a prototype-based shared-dummy classifier (PSDC) model for OSDA.
• Developed an auxiliary dummy classifier to calibrate the source classifier.
• Implemented a weighted adaptation procedure for aligning class-wise prototypes and a pseudo-unknown learning algorithm.

Main Results:

• The proposed PSDC model demonstrated superior performance compared to existing methods.
• Achieved new state-of-the-art results on benchmark datasets including Office-31, Office-Home, and VisDA.
• Effectively addressed the limitations of treating unknown instances as a single group and improved class-wise adaptation.

Conclusions:

• The PSDC model offers a significant advancement in open-set domain adaptation.
• The method's ability to align class-wise prototypes and handle unknown classes leads to improved performance.
• The proposed approach sets a new standard for OSDA, with code availability to facilitate further research.