Generalized SMO algorithm for SVM-based multitask learning.

Area of Science:

• Machine Learning
• Computational Learning Theory

Background:

• Supervised learning can benefit from incorporating group information in training data.
• Support Vector Machine plus (SVM+) and Multitask Learning (MTL) are advanced machine learning approaches.
• Existing SVM+MTL classifiers face computational challenges due to large-scale optimization problems.

Purpose of the Study:

• To develop a computationally efficient algorithm for training SVM+MTL classifiers.
• To generalize the Sequential Minimal Optimization (SMO) algorithm for the SVM+MTL framework.
• To improve the practical applicability of structured data learning methods.

Main Methods:

• Generalization of Platt's Sequential Minimal Optimization (SMO) algorithm.
• Application to Support Vector Machine plus Multitask Learning (SVM+MTL) classification.
• Comparative analysis against general-purpose optimization routines.

Main Results:

• The proposed generalized SMO algorithm significantly reduces computational time for SVM+MTL.
• Achieved over 100 times speed-up compared to standard optimization methods for typical problems.
• Demonstrated the efficiency of the generalized SMO for structured data learning.

Conclusions:

• The generalized SMO algorithm provides a computationally efficient solution for SVM+MTL.
• This advancement makes structured data learning more accessible and practical.
• The findings contribute to improving generalization in machine learning through structured data exploitation.