A novel ramp loss-based multi-task twin support vector machine with multi-parameter safe acceleration.

Area of Science:

• Machine Learning
• Data Mining
• Optimization

Background:

• Direct multi-task support vector machines (DMTSVM) use hinge loss, making them sensitive to outliers.
• Existing methods struggle with noisy data in multi-task classification.

Purpose of the Study:

• To develop a robust multi-task classification model less affected by noisy data.
• To accelerate the computation of a non-convex multi-task learning model.

Main Methods:

• Proposed a ramp loss-based multi-task support vector machine (RaMTTSVM) to mitigate outlier influence.
• Introduced a safe acceleration rule (MSA) based on optimality conditions and convex optimization theory to speed up the CCCP algorithm.
• MSA safely removes inactive samples before solving, reducing problem size without impacting optimal solutions.

Main Results:

• RaMTTSVM demonstrates improved robustness against noisy data compared to DMTSVM.
• The MSA rule significantly accelerates the solving speed of RaMTTSVM.
• Experimental results across diverse datasets confirm the model's generalization and acceleration capabilities.

Conclusions:

• The proposed RaMTTSVM with MSA offers an effective and efficient solution for multi-task classification with noisy data.
• The safe acceleration rule ensures computational efficiency without compromising prediction accuracy.