Steps:

1. Get initial guess($prediction$) by taking the average of the dataset
2. Get $residual\mathrm{\_}output=ground\mathrm{\_}truth-guess$
3. Build a tree to fit the $residual\mathrm{\_}output$
4. Now the prediction will be $prediction=prediction+lr\ast new\mathrm{\_}tree\mathrm{\_}prediction$
   1. remember this $new\mathrm{\_}tree\mathrm{\_}prediction$ is predicted residual
5. Go to step 2 unless predetermined limit is reached

The regression case of Gradient Boosting (see Gradient Boost (Classification) and AdaBoost vs. Gradient Boosting vs. XGBoost); each step fits a Decision Tree to the residual.