Incorporating Causal Prior Knowledge into Deep Neural Networks

Deep Neural Networks have achieved significant success in solving complex problems across various domains due to their ability to capture complicated patterns in large datasets; however, they often require large amounts of data to learn effectively and often lack transparency in their decision-making processes, relying heavily on correlation rather than causation. Such limitations have led
to incorporating causal Prior Knowledge into neural network models which stands as a significant advancement in machine learning, such knowledge can mitigate this data dependency, guide the learning process, and enhance not only the robustness and generalizability of models but also their interpretability and explainability. Additionally, it enables models to adapt to new tasks and domains with greater ease and effectiveness.

This report tackles the importance of incorporating causal prior knowledge into deep neural networks and the methodologies that facilitate this incorporation. Fundamental concepts of causality are reviewed, with emphasis on its importance for advancing AI towards causal representation learning.