Algorithm 1.

KbPIB: Knowledge-based Privileged Information Boosting.

Input: Classifier features: training data $X_{\text{train}}^{\text{C}\text{F}},Y_{\text{train}}$; validation data $X_{\text{val}}^{\text{C}\text{F}},Y_{\text{val}}$; privileged features: training data $X_{\text{train}}^{\text{P}\text{F}},Y_{\text{train}}$; validation data $X_{\text{val}}^{\text{P}\text{F}},Y_{\text{val}}$

Parameter: Number of trees N, early-stop parameter POutput: Learned model ψ

1: Initialize model ψ0 = 0, counter C = 0, score R, best number of trees index j

2: ψPF← NF($X_{\text{train}}^{\text{P}\text{F}},Y_{\text{train}},X_{\text{val}}^{\text{P}\text{F}},Y_{\text{val}}$) { Supplementary Algorithm 1}

3: for i = 1 to N do

4: Δi← ComputeGradient($X_{\text{train}}^{\text{C}\text{F}},Y_{\text{train}},{\psi }_{i-1},{\psi }^{\text{P}\text{F}}$) {Equation (2)}

5: ${\widehat{\Delta }}_i\leftarrow$ FitRegressionValue($X_{\text{train}}^{\text{C}\text{F}},{\Delta }_i$)

6: ${\psi }_i\leftarrow {\psi }_{i-1}+{\widehat{\Delta }}_i$

7: Rval ← Evaluate($X_{\text{val}}^{\text{C}\text{F}},Y_{\text{val}},{\psi }_i$)

8: j, R, C ← EarlyStop(i, j, R, Rval, C, P) { Supplementary Algorithm 2}

9: end for

10: return ψj