Importance Weighted Import Vector Machine for Unsupervised Domain Adaptation.

Area of Science:

• Machine Learning
• Computer Vision
• Artificial Intelligence

Background:

• Real-world data often violates the assumption of independent and identical distribution between source and target domains.
• This mismatch significantly degrades the performance of standard machine learning models in cross-domain applications.

Purpose of the Study:

• To propose a novel adaptive classifier for unsupervised domain adaptation (DA) that effectively handles distribution mismatches.
• To introduce a robust cross-validation technique for improved parameter and model selection in DA.

Main Methods:

• Developed an importance weighting import vector machine (IVM) as a sparse and computationally efficient adaptive classifier.
• Applied the IVM to unsupervised domain adaptation (DA) scenarios.
• Introduced Reliable Importance Weighted Cross-Validation (RIWCV) to enhance parameter and model selection, avoiding local minima.

Main Results:

• The proposed importance weighting IVM demonstrated superior performance compared to state-of-the-art methods in both unsupervised and semi-supervised DA.
• Effectiveness was validated on a toy problem and a real-world cross-domain object recognition task.
• RIWCV successfully identified more reliable parameter combinations, improving model selection.

Conclusions:

• The importance weighting IVM is a computationally efficient and effective solution for unsupervised domain adaptation.
• The proposed RIWCV offers a more reliable approach to parameter and model selection, outperforming traditional methods.