Multi-source transfer learning for facial emotion recognition using multivariate correlation analysis.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Facial emotion recognition (FER) is crucial for human-computer interaction.
• Deep learning models for FER require substantial labeled data, which is often limited by privacy and ethical concerns.
• Existing methods struggle with data scarcity and negative transfer in multi-source learning.

Purpose of the Study:

• To develop a novel multi-source transfer learning approach for facial emotion recognition (FER).
• To address the challenge of limited labeled data in FER by leveraging knowledge from multiple related data sources.
• To improve the robustness and performance of FER models in few-shot learning scenarios.

Main Methods:

• Proposed a multi-source transfer learning framework for FER.
• Optimized aggregate multivariate correlation among source tasks to control information transfer.
• Validated the approach on benchmark datasets for FER and image classification.

Main Results:

• The method effectively captures group correlations among features.
• Demonstrated robustness against negative transfer.
• Achieved strong performance in few-shot multi-source adaptation.
• Outperformed state-of-the-art methods MCW and DECISION by 7% and 15% respectively.

Conclusions:

• The proposed multi-source transfer learning approach is effective for FER, especially with limited data.
• The method offers a robust solution to negative transfer challenges in transfer learning.
• This technique shows significant potential for advancing FER and related machine learning applications.