Experience-consistent modeling for radial basis function neural networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Traditional neural network design relies heavily on experimental data.
• Integrating prior knowledge into neural network construction is challenging.
• Existing methods often overlook valuable information embedded in previously trained network parameters.

Purpose of the Study:

• To develop a novel, knowledge-driven approach for designing neural networks.
• To create a framework that reconciles experimental data with prior knowledge.
• To focus on experience-based design of radial basis function neural networks (RBFNNs).

Main Methods:

• A collaborative framework integrating knowledge-driven experience with current data.
• Development of a conceptual and algorithmic framework for information reconciliation.
• Concentration on radial basis function neural networks (RBFNNs) for focused quantification.
• Introduction of performance indexes to quantify the impact of utilizing existing network knowledge.

Main Results:

• Demonstrated a method to effectively utilize past neural network parameters (knowledge) alongside new data.
• Quantified the impact of knowledge integration on RBFNN architecture and performance.
• Established optimal levels for the utilization of embedded knowledge.
• Validated the approach using synthetic and real-world datasets.

Conclusions:

• The proposed knowledge-driven experience approach offers a powerful alternative to standard data-driven neural network design.
• Integrating prior knowledge significantly enhances the development and performance of neural networks, particularly RBFNNs.
• The developed framework and performance indexes provide a robust method for quantifying and optimizing knowledge utilization in AI systems.