Active Multitask Learning With Trace Norm Regularization Based on Excess Risk.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Labeled data is costly to acquire for individual tasks.
• Related tasks often share commonalities that can be leveraged.
• Active learning aims to reduce labeling costs by selecting informative data points.

Purpose of the Study:

• To develop a novel active multitask learning (AML) approach.
• To address the challenge of expensive data labeling across multiple related tasks.
• To improve the efficiency and performance of active learning.

Main Methods:

• Proposed a trace norm regularized least squares method for AML.
• Devised a new active selection criterion considering both risk and excess risk.
• Implemented an algorithm that actively selects instances based on combined risk assessment.

Main Results:

• The proposed AML algorithm demonstrated superior performance.
• Achieved better results compared to existing state-of-the-art active learning methods.
• Validated through experiments on both synthetic and real-world datasets.

Conclusions:

• The novel AML approach effectively leverages shared information across tasks.
• The proposed active selection criterion enhances labeling efficiency.
• This method offers a promising solution for cost-effective data acquisition in multitask learning.