KE: A Knowledge Enhancing Framework for Machine Learning Models

Area of Science:

• Materials Science
• Computer Science
• Artificial Intelligence

Background:

• Machine learning models are crucial for predicting material properties but suffer from lengthy training and hyperparameter tuning times.
• Achieving optimal performance in complex scientific and engineering problems using large models remains a significant challenge.

Purpose of the Study:

• To develop a novel Knowledge Enhancing (KE) algorithm to improve the efficiency and performance of machine learning model training.
• To address the computational challenges associated with training large-scale models in scientific applications.

Main Methods:

• The proposed Knowledge Enhancing (KE) algorithm transfers knowledge from a lower-capacity model to a higher-capacity model.
• The algorithm's effectiveness is demonstrated through theoretical analysis and experimental verification, focusing on predicting material bandgaps.
• Experiments were conducted using the OMDB dataset to evaluate performance against existing methods.

Main Results:

• The Knowledge Enhancing (KE) model demonstrated a performance improvement of at least 10.21% compared to current methods on OMDB datasets.
• The algorithm successfully enhanced knowledge transfer, leading to improved training efficiency and predictive accuracy for material bandgaps.

Conclusions:

• The Knowledge Enhancing (KE) algorithm offers a promising approach to accelerate machine learning model training and enhance performance in scientific discovery.
• The generic nature of knowledge enhancement suggests broad applicability to various scientific and engineering problems, paving the way for future research.