Multi-Task Learning Based on Stochastic Configuration Networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Human brain exhibits knowledge sharing and transfer when learning multiple related or continuous tasks.
• This phenomenon inspires multi-task learning (MTL) approaches for efficient model development.
• Effective MTL relies on identifying task correlations and building models based on this information.

Purpose of the Study:

• To propose a novel multi-task learning framework leveraging stochastic configuration networks (SCNs).
• To integrate classical parameter sharing in MTL with constraint sharing configurations in SCNs.
• To introduce an efficient multi-kernel function selection mechanism.

Main Methods:

• Developed a multi-task learning framework based on stochastic configuration networks.
• Combined parameter sharing from classical MTL with constraint sharing from SCNs.
• Implemented an efficient multi-kernel function selection mechanism.

Main Results:

• Theoretically proved the convergence of the proposed algorithm.
• Experimental validation on one simulation and four real-world datasets demonstrated effectiveness.
• The framework successfully facilitates knowledge sharing and transfer between tasks.

Conclusions:

• The proposed SCN-based MTL framework is effective for learning multiple related tasks.
• The integration of parameter and constraint sharing enhances learning efficiency.
• The method offers a promising approach for complex learning scenarios.