Update Neural_Turing_Machines_Reports_and_Discussion.ipynb

Author: rezazzr

Neural_Turing_Machines_Reports_and_Discussion.ipynb

@@ -208,7 +208,7 @@
       "source": [
         "### **(b)** Hyper-parameter Values\n",
         "We experimented with a large set of candidates to choose our hyper-parameters from. We found out a few suitable candidates for our tasks which worked well across all our methods. Below we are providing the hyper-parameters we eventually used.\n",
-        "<center><table style=\"width:100%\" border=\"1px solid black\">\n",
+        "<table style=\"width:100%\" border=\"1px solid black\">\n",
         "  <tr>\n",
         "    <th>Method's Name</th>\n",
         "    <th>Learning Rate</th>\n",
@@ -219,33 +219,33 @@
         "    <th>Gradien Clipped at Max Norm of</th>\n",
         "  </tr>\n",
         "  <tr>\n",
-        "    <td><center>LSTM-NTM</center></td>\n",
-        "    <td><center>0.01</center></td>\n",
-        "    <td><center>0.5</center></td>\n",
-        "    <td><center>100</center></td>\n",
-        "    <td><center>Stochastic Gradient Descent</center></td>\n",
-        "    <td><center>Softmax</center></td>\n",
-        "    <td><center>10</center></td>\n",
+        "    <td>LSTM-NTM</td>\n",
+        "    <td>0.01</td>\n",
+        "    <td>0.5</td>\n",
+        "    <td>100</td>\n",
+        "    <td>Stochastic Gradient Descent</td>\n",
+        "    <td>Softmax</td>\n",
+        "    <td>10</td>\n",
         "  </tr>\n",
         "  <tr>\n",
-        "    <td><center>Feedforward-NTM</center></td>\n",
-        "    <td><center>0.01</center></td>\n",
-        "    <td><center>0.5</center></td>\n",
-        "    <td><center>100</center></td>\n",
-        "    <td><center>Stochastic Gradient Descent</center></td>\n",
-        "    <td><center>Softmax</center></td>\n",
-        "    <td><center>10</center></td>\n",
+        "    <td>Feedforward-NTM</td>\n",
+        "    <td>0.01</td>\n",
+        "    <td>0.5</td>\n",
+        "    <td>100</td>\n",
+        "    <td>Stochastic Gradient Descent</td>\n",
+        "    <td>Softmax</td>\n",
+        "    <td>10</td>\n",
         "  </tr>\n",
         "  <tr>\n",
-        "    <td><center>LSTM</center></td>\n",
-        "    <td><center>0.001</center></td>\n",
-        "    <td><center>0.5</center></td>\n",
-        "    <td><center>100</center></td>\n",
-        "    <td><center>Adam</center></td>\n",
-        "    <td><center>Softmax</center></td>\n",
-        "    <td><center>10</center></td>\n",
+        "    <td>LSTM</td>\n",
+        "    <td>0.001</td>\n",
+        "    <td>0.5</td>\n",
+        "    <td>100</td>\n",
+        "    <td>Adam</td>\n",
+        "    <td>Softmax</td>\n",
+        "    <td>10</td>\n",
         "  </tr>\n",
-        "</table></center><br>\n",
+        "</table><br>\n",
         "* Note: In order to eliminate other hyper-parameter values and to pick the best ones for our tasks, we only let each model to progress for $200$ epochs, we observed that the loss curve decreases towards $0$. We then picked the values for which the loss curve better approached zero. The main reason we did not let our models to be trained passed $200$ epochs was purely due to our limited computational resources. We will discuss these issues further in the **Discussion** section."
       ]
     },