A new learning paradigm for random vector functional-link network: RVFL.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Neural Networks

Background:

• The learning using privileged information (LUPI) paradigm mimics human teacher-student interaction by providing extra training data.
• Random Vector Functional Link (RVFL) networks are a type of randomized neural network.
• There was a gap in integrating LUPI with classical randomized neural networks.

Purpose of the Study:

• To introduce the first RVFL network, termed RVFL+, that incorporates the LUPI paradigm.
• To enhance RVFL training by leveraging additional information sources.
• To investigate the statistical properties and generalization error bounds of RVFL+.

Main Methods:

• Developed RVFL+, a novel RVFL network architecture integrating the LUPI paradigm.
• Introduced KRVFL+, an extension of RVFL+ that combines with the kernel trick for nonlinear feature learning.
• Analyzed the statistical properties and derived generalization error bounds using Rademacher complexity.

Main Results:

• RVFL+ offers an alternative training method for RVFL networks.
• KRVFL+ effectively handles highly complicated nonlinear feature learning.
• Experiments on 14 real-world datasets demonstrated superior generalization performance compared to state-of-the-art methods.

Conclusions:

• The proposed RVFL+ and KRVFL+ models are effective and efficient for machine learning tasks.
• These novel approaches bridge the gap between classical randomized networks and the LUPI paradigm.
• The methods achieve improved generalization, showcasing their practical utility.