Causal Inference Meets Deep Learning: A Comprehensive Survey.

Area of Science:

• Artificial Intelligence
• Cognitive Neuroscience
• Machine Learning

Background:

• Deep learning models learn from data but can capture spurious correlations, limiting interpretability and robustness.
• Traditional correlation-based models in deep learning face challenges with misleading data patterns.

Purpose of the Study:

• To provide a comprehensive review of causal inference methods for deep learning.
• To explore brain-inspired approaches for enhancing deep learning model stability and interpretability.

Main Methods:

• Reviewing causal inference techniques integrated with deep learning algorithms.
• Discussing brain-like inference principles and causal learning concepts.
• Examining applications in large model tasks and specific deep learning modalities.

Main Results:

• Causal inference offers a path to mitigate spurious correlations in deep learning.
• Brain-inspired methods can lead to more stable and interpretable AI models.
• Integration of causal inference enhances deep learning performance and reliability.

Conclusions:

• Causal inference is crucial for advancing robust and interpretable deep learning.
• Future research should focus on refining causal inference techniques and their applications.
• This survey provides a structured overview and resources for causal deep learning research.