Integrated Gradients
An explainable AI technique that attributes a model's prediction to its input features by integrating gradients along a path from a baseline to the input.
Think of It Like This
Like tracing the exact flow of water from a mountain spring down into a valley to understand which streams contributed the most.
This method satisfies important axioms like sensitivity and implementation invariance, making it robust for interpreting deep neural networks. It computes the feature importance by accumulating the gradients of the model's output with respect to the input as it scales. It is widely used to visualize which pixels or words drove a specific prediction.