add l2 to Conv.

luma/__import__.py

@@ -75,7 +75,7 @@
 )
 from luma.neural.single import PerceptronClassifier, PerceptronRegressor
 from luma.neural.network import MLPClassifier, MLPRegressor
-from luma.neural.layer import Convolution, Pooling, Dense
+from luma.neural.layer import Convolution, Pooling, Dense, Dropout, Flatten
 
 from luma.metric.classification import Accuracy, Precision, Recall, F1Score, Specificity
 from luma.metric.regression import (
@@ -223,7 +223,7 @@
     AdamOptimizer, AdaGradOptimizer, AdaDeltaOptimizer,
     AdaMaxOptimizer, AdamWOptimizer, NAdamOptimizer
 
-    Convolution, Pooling, Dense
+    Convolution, Pooling, Dense, Dropout, Flatten
 
     # ------------------- [ luma.metric ] ----------------------
     Accuracy, Precision, Recall, F1Score, Specificity

luma/neural/layer.py

@@ -27,6 +27,7 @@ class Convolution(Layer):
     (`valid` for no padding, `same` for typical 0-padding)
     `activation` : Type of activation function
     `optimizer` : Optimizer for weight update (default `SGDOptimizer`)
+    `lambda_` : L2-regularization strength
     `random_state` : Seed for various random sampling processes
 
     Notes
@@ -46,6 +47,7 @@ def __init__(
         padding: Literal["valid", "same"] = "same",
         activation: ActivationUtil.FuncType = "relu",
         optimizer: Optimizer = SGDOptimizer(),
+        lambda_: float = 0.1,
         random_state: int = None,
     ) -> None:
         super().__init__()
@@ -55,6 +57,7 @@ def __init__(
         self.padding = padding
         self.activation = activation
         self.optimizer = optimizer
+        self.lambda_ = lambda_
 
         act = ActivationUtil(self.activation)
         self.act_ = act.activation_type()
@@ -125,6 +128,8 @@ def backward(self, d_out: Tensor) -> Tuple[Tensor, Tensor, Tensor]:
                     filter_d_out_fft, s=(padded_height, padded_width)
                 )[pad_h : pad_h + self.size, pad_w : pad_w + self.size]
 
+        self.dW += 2 * self.lambda_ * self.weights_
+
         for i in range(batch_size):
             for c in range(channels):
                 temp = np.zeros(