Active Learning for Multiple Target Models.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Traditional active learning (AL) methods are often model-dependent and lack transferability.
• Real-world applications frequently require training multiple models for diverse computational resources.
• Existing AL approaches face challenges in multi-model learning scenarios.

Purpose of the Study:

• To investigate the feasibility of designing effective active learning methods for simultaneously learning multiple target models.
• To analyze the query complexity of active versus passive learning in a multi-model setting.
• To develop a novel AL strategy applicable across diverse machine learning models.

Main Methods:

• Analysis of query complexity for active and passive learning in the multi-model setting.
• Proposal of an agnostic active learning sampling strategy.
• Selection of data points from joint disagreement regions among different target models.

Main Results:

• Demonstrated potential for active learning to achieve improved query complexity in multi-model learning.
• Validated the effectiveness of the proposed agnostic AL sampling strategy.
• Experimental results show superior performance compared to traditional AL methods on benchmark datasets.

Conclusions:

• An effective active learning method can be designed for simultaneously learning multiple target models.
• The proposed agnostic AL strategy offers a promising direction for efficient multi-model training.
• This approach enhances data efficiency in machine learning systems with diverse model requirements.