Fix some typos in the Uncertainty-aware Deep Learning with SNGP tutorial colab.

PiperOrigin-RevId: 379287145

Author: tensorflower-gardener

site/en/tutorials/understanding/sngp.ipynb

@@ -499,7 +499,7 @@
       },
       "outputs": [],
       "source": [
-        "resnet_model.compile(optimizer=optimizer, loss=loss, metrics=metrics)\n",
+        "resnet_model.compile(**train_config)\n",
         "resnet_model.fit(train_examples, train_labels, **fit_config)"
       ]
     },
@@ -838,7 +838,7 @@
         "id": "P13X7Adt-c2d"
       },
       "source": [
-        "Note: The momentum-based update method can be sensitive to batch size. Therefore it is generally recommended to set  `gp_cov_momentum=-1` to compute the covariance exactly. For this to work properly, the covariance matrix estimator needs to be reset at the begining of a new epoch in order to avoid counting the same data twice. For `RandomFeatureGaussianProcess`, this is can be done by calling its `rest_covariance_matrix()`. The next section shows an easy implementation of this using Keras' built-in API.\n"
+        "Note: The momentum-based update method can be sensitive to batch size. Therefore it is generally recommended to set  `gp_cov_momentum=-1` to compute the covariance exactly. For this to work properly, the covariance matrix estimator needs to be reset at the begining of a new epoch in order to avoid counting the same data twice. For `RandomFeatureGaussianProcess`, this is can be done by calling its `reset_covariance_matrix()`. The next section shows an easy implementation of this using Keras' built-in API.\n"
       ]
     },
     {
@@ -871,7 +871,7 @@
       "source": [
         "Note: Notice that under this implementation of the SNGP model, the predictive logits $logit(x_{test})$ for all classes share the same covariance matrix $var(x_{test})$, which describes the distance between $x_{test}$ from the training data. \n",
         "\n",
-        "Theoretically, it is possible to extend the algorithm to compute different variance values for different classes (as introduced in the [original SNGP paper](https://arxiv.org/abs/2006.10108)). However, this is diffcult to scale to problems with large output spaces (e.g., ImageNet or language modeling)."
+        "Theoretically, it is possible to extend the algorithm to compute different variance values for different classes (as introduced in the [original SNGP paper](https://arxiv.org/abs/2006.10108)). However, this is difficult to scale to problems with large output spaces (e.g., ImageNet or language modeling)."
       ]
     },
     {
@@ -1512,7 +1512,7 @@
         "In this tutorial, you have:\n",
         "* Implemented a SNGP model on a deep classifier to improve its distance awareness.\n",
         "* Trained the SNGP model end-to-end using Keras `model.fit()` API.\n",
-        "* Visualized the uncertainty behavior of SNGP\n",
+        "* Visualized the uncertainty behavior of SNGP.\n",
         "* Compared the uncertainty behavior between SNGP, Monte Carlo dropout and deep ensemble models."
       ]
     },
@@ -1531,7 +1531,7 @@
         "id": "HoIikRybke-b"
       },
       "source": [
-        "* See the [SNGP-BERT tutorial](https://www.tensorflow.org/official_models/tutorials/uncertainty_quantification_with_sngp_bert) for an example of applying SNGP on a BERT model for uncertainty-aware natural language understanding. \n",
+        "* See the [SNGP-BERT tutorial](https://www.tensorflow.org/text/tutorials/uncertainty_quantification_with_sngp_bert) for an example of applying SNGP on a BERT model for uncertainty-aware natural language understanding.\n",
         "* See [Uncertainty Baselines](https://github.com/google/uncertainty-baselines)  for the implementation of SNGP model (and many other uncertainty methods) on a wide variety of benchmark datasets (e.g., [CIFAR](https://www.tensorflow.org/datasets/catalog/cifar100), [ImageNet](https://www.tensorflow.org/datasets/catalog/imagenet2012), [Jigsaw toxicity detection](https://www.tensorflow.org/datasets/catalog/wikipedia_toxicity_subtypes), etc).\n",
         "* For a deeper understanding of the SNGP method, check out the paper [Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness](https://arxiv.org/abs/2006.10108).\n"
       ]