Multi-Modality Fusion & Inductive Knowledge Transfer Underlying Non-Sparse Multi-Kernel Learning and Distribution...

Area of Science:

• Biomedical data analysis
• Bioinformatics
• Machine learning

Background:

• Increasing accumulation of multimodal data in biomedical and bioinformatics fields necessitates advanced analysis techniques.
• Multimodal data analysis is crucial for extracting comprehensive insights but is often hindered by insufficient training samples.

Purpose of the Study:

• To propose a novel feature-level multi-modality fusion model designed for situations with insufficient training samples.
• To enhance the exploration of complementary patterns within multimodal data using advanced machine learning techniques.

Main Methods:

• Extension of kernel Ridge regression to a multi-kernel version with an lp-norm constraint for multimodal data pattern exploration.
• Application of marginal probability distribution adaptation to minimize domain discrepancies and overcome limited training data challenges.
• Evaluation using epilepsy electroencephalogram (EEG) data across 12 multi-modality and transfer learning scenarios.

Main Results:

• The proposed model demonstrated superior performance compared to baseline methods across the majority of evaluated scenarios.
• Effective fusion of multimodal data was achieved, even with limited available training samples.
• The model successfully leveraged transfer learning principles to improve analytical outcomes.

Conclusions:

• The developed multi-kernel and transfer learning fusion model offers a robust solution for multimodal data analysis with insufficient training samples.
• The approach shows significant potential for applications in biomedical and bioinformatics research, particularly with complex datasets like EEG.
• Further validation across diverse multimodal datasets is recommended to establish broader applicability.