diff --git a/mindarmour/privacy/diff_privacy/mechanisms/mechanisms.py b/mindarmour/privacy/diff_privacy/mechanisms/mechanisms.py
index 61d755f..bfb33cb 100644
--- a/mindarmour/privacy/diff_privacy/mechanisms/mechanisms.py
+++ b/mindarmour/privacy/diff_privacy/mechanisms/mechanisms.py
@@ -284,6 +284,21 @@ class NoiseAdaGaussianRandom(NoiseGaussianRandom):
                             "get {}".format(decay_policy))
         self._decay_policy = decay_policy
 
+    def construct(self, gradients):
+        """
+        Generated Adaptive Gaussian noise.
+
+        Args:
+            gradients(Tensor): The gradients.
+
+        Returns:
+            Tensor, generated noise with shape like given gradients.
+        """
+        shape = P.Shape()(gradients)
+        stddev = P.Mul()(self._norm_bound, self._noise_multiplier)
+        noise = normal(shape, self._mean, stddev, self._seed)
+        return noise
+
 
 class _MechanismsParamsUpdater(Cell):
     """
diff --git a/mindarmour/privacy/diff_privacy/train/model.py b/mindarmour/privacy/diff_privacy/train/model.py
index 805dd4e..5454e63 100644
--- a/mindarmour/privacy/diff_privacy/train/model.py
+++ b/mindarmour/privacy/diff_privacy/train/model.py
@@ -515,7 +515,7 @@ class _TrainOneStepWithLossScaleCell(Cell):
         if self._noise_mech is not None:
             grad_noise_tuple = ()
             for grad_item in grads:
-                grad_noise = self._mech(grad_item)
+                grad_noise = self._noise_mech(grad_item)
                 grad_noise_tuple = grad_noise_tuple + (grad_noise,)
             grads = self._tuple_add(grads, grad_noise_tuple)
             grads = self._hyper_map(F.partial(_grad_scale, self._micro_float),