diff --git a/keras/backend/mxnet_backend.py b/keras/backend/mxnet_backend.py
index 206121bd7533..fb0910cfa3f5 100644
--- a/keras/backend/mxnet_backend.py
+++ b/keras/backend/mxnet_backend.py
@@ -203,7 +203,11 @@ def __init__(self, symbol, name=None, neighbor=None, is_var=False):
 
 
     def bind(self, data):
-        self.tensor = data
+        if not hasattr(self, 'tensor'):
+            self.tensor = data
+        else:
+            self.tensor[:] = data
+            
         if self.name in self._bind_values:
             assert self._bind_values[self.name].shape == data.shape, \
                 "Redefinition of variable %s" % self.name
@@ -884,7 +888,8 @@ def random_uniform_variable(shape, low, high, dtype=None,
     value = mx.random.uniform(low=low, high=high, dtype='float32', shape=shape)
     if dtype != np.float32:
         value = mx.nd.Cast(value, dtype=dtype)
-    name = _autogen_name("uniform")
+    if name is None:
+        name = _autogen_name("uniform")
     ret = KerasVariable(name, value.shape, value.dtype)
     ret.bind(value)
     return ret
@@ -923,7 +928,8 @@ def random_normal_variable(shape, mean, scale, dtype=None,
     value = mx.random.normal(loc=mean, scale=scale, dtype='float32', shape=shape)
     if dtype != np.float32:
         value = mx.nd.Cast(value, dtype=dtype)
-    name = _autogen_name("normal")
+    if name is None:
+        name = _autogen_name("normal")
     ret = KerasVariable(name, value.shape, value.dtype)
     ret.bind(value)
     return ret
@@ -2956,7 +2962,8 @@ def random_normal(shape, mean=0.0, std=1.0, dtype=None, seed=None):
     if dtype is None:
         dtype = floatx()
     dtype = np.dtype(dtype)
-    name = _autogen_name('normal')
+    if name is None:
+        name = _autogen_name('normal')
     sym = mx.sym.normal(loc=mean, scale=std, shape=shape, dtype='float32', name=name)
     if dtype != np.float32:
         sym = mx.sym.Cast(data=sym, dtype=dtype)
@@ -2983,7 +2990,8 @@ def random_uniform(shape, low=0.0, high=1.0, dtype=None, seed=None):
     if dtype is None:
         dtype = floatx()
     dtype = np.dtype(dtype)
-    name = _autogen_name('uniform')
+    if name is None:
+        name = _autogen_name('uniform')
     sym = mx.sym.uniform(low=low, high=high, shape=shape, dtype='float32', name=name)
     if dtype != np.float32:
         sym = mx.sym.Cast(data=sym, dtype=dtype)