Active Self-Paced Learning for Cost-Effective and Progressive Face Identification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Face identification systems require extensive annotated data for training.
• Existing methods struggle with efficiency and robustness against noisy data.

Purpose of the Study:

• To develop a novel, cost-effective framework for face identification.
• To reduce the need for annotated samples and user effort.
• To enhance classifier accuracy and robustness.

Main Methods:

• Combines active learning (AL) and self-paced learning (SPL) for automatic annotation and training.
• Utilizes convolutional neural networks for feature extraction.
• Implements an active SPL optimization problem with a dynamic curriculum constraint.

Main Results:

• Significantly decreases the number of required annotated samples while maintaining performance.
• Achieves a dramatic reduction in user effort compared to state-of-the-art AL techniques.
• Improves classifier accuracy and robustness against noisy data.

Conclusions:

• The proposed framework offers a cost-effective and efficient solution for face identification.
• The integration of AL and SPL enhances model performance and resilience.
• The approach aligns with collaborative learning paradigms and shows promising results on challenging datasets.