Gradient is the connection line that enables causal interaction between different components, and Residual Connection enables this global causal connection without spatial restriction.
Gradient Interpretability Methods
Tracking dataset influence by remove noisy datapoints
Estimating Training Data Influence by Tracing Gradient Descent
We introduce a method called TracIn that computes the influence of a training example on a prediction made by the model. The idea is to trace how the loss on the test point changes during the...
https://arxiv.org/abs/2002.08484

Filter for the data to finetune based on gradient

Seonglae Cho