| -- SVM Geometric Margin Maximization Body. | |
| package body SVM_Margin is | |
| procedure Optimize_Hyperplane ( | |
| Features : in Vector; | |
| Label : in Float_32; | |
| Weights : in out Vector; | |
| Bias : in out Float_32; | |
| Learning_Rate : in Float_32; | |
| Lambda : in Float_32 | |
| ) is | |
| Dot_Product : Float_32 := 0.0; | |
| Margin_Test : Float_32; | |
| begin | |
| -- 1. Calculate the geometric projection: W^T * X | |
| for I in Features'Range loop | |
| Dot_Product := Dot_Product + (Weights(I) * Features(I)); | |
| end loop; | |
| -- 2. Test the functional margin: y_i * (W^T * X_i + b) | |
| Margin_Test := Label * (Dot_Product + Bias); | |
| -- 3. Hinge Loss Gradient Update | |
| if Margin_Test >= 1.0 then | |
| -- Vector is outside the margin boundary. | |
| -- Apply weight decay (L2 regularization) to maximize geometric margin width. | |
| for I in Weights'Range loop | |
| Weights(I) := Weights(I) - Learning_Rate * (Lambda * Weights(I)); | |
| end loop; | |
| else | |
| -- Vector violates the margin boundary (Support Vector). | |
| -- Adjust the hyperplane geometry toward the structural invariant. | |
| for I in Weights'Range loop | |
| Weights(I) := Weights(I) - Learning_Rate * (Lambda * Weights(I) - Label * Features(I)); | |
| end loop; | |
| -- Mutate the bias threshold | |
| Bias := Bias - Learning_Rate * (-Label); | |
| end if; | |
| end Optimize_Hyperplane; | |
| end SVM_Margin; | |