Fixed the errors in gradient_centralization.py (#2042)

* Update gradient_centralization.py

* updated

Author: tilakrayal

examples/vision/gradient_centralization.py

@@ -151,7 +151,7 @@ def prepare(ds, shuffle=False, augment=False):
 subclass the `RMSProp` optimizer class modifying the
 `keras.optimizers.Optimizer.get_gradients()` method where we now implement Gradient
 Centralization. On a high level the idea is that let us say we obtain our gradients
-through back propogation for a Dense or Convolution layer we then compute the mean of the
+through back propagation for a Dense or Convolution layer we then compute the mean of the
 column vectors of the weight matrix, and then remove the mean from each column vector.
 
 The experiments in [this paper](https://arxiv.org/abs/2004.01461) on various

examples/vision/ipynb/gradient_centralization.ipynb

@@ -64,8 +64,7 @@
     "from keras import ops\n",
     "\n",
     "from tensorflow import data as tf_data\n",
-    "import tensorflow_datasets as tfds\n",
-    ""
+    "import tensorflow_datasets as tfds\n"
    ]
   },
   {
@@ -159,8 +158,7 @@
     "        )\n",
     "\n",
     "    # Use buffered prefecting\n",
-    "    return ds.prefetch(buffer_size=AUTOTUNE)\n",
-    ""
+    "    return ds.prefetch(buffer_size=AUTOTUNE)\n"
    ]
   },
   {
@@ -238,7 +236,7 @@
     "subclass the `RMSProp` optimizer class modifying the\n",
     "`keras.optimizers.Optimizer.get_gradients()` method where we now implement Gradient\n",
     "Centralization. On a high level the idea is that let us say we obtain our gradients\n",
-    "through back propogation for a Dense or Convolution layer we then compute the mean of the\n",
+    "through back propagation for a Dense or Convolution layer we then compute the mean of the\n",
     "column vectors of the weight matrix, and then remove the mean from each column vector.\n",
     "\n",
     "The experiments in [this paper](https://arxiv.org/abs/2004.01461) on various\n",
@@ -314,8 +312,7 @@
     "        self.epoch_time_start = time()\n",
     "\n",
     "    def on_epoch_end(self, batch, logs={}):\n",
-    "        self.times.append(time() - self.epoch_time_start)\n",
-    ""
+    "        self.times.append(time() - self.epoch_time_start)\n"
    ]
   },
   {
@@ -473,4 +470,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 0
-}
+}

examples/vision/md/gradient_centralization.md

@@ -170,7 +170,7 @@ We will now
 subclass the `RMSProp` optimizer class modifying the
 `keras.optimizers.Optimizer.get_gradients()` method where we now implement Gradient
 Centralization. On a high level the idea is that let us say we obtain our gradients
-through back propogation for a Dense or Convolution layer we then compute the mean of the
+through back propagation for a Dense or Convolution layer we then compute the mean of the
 column vectors of the weight matrix, and then remove the mean from each column vector.
 
 The experiments in [this paper](https://arxiv.org/abs/2004.01461) on various