update

README.md

@@ -5,14 +5,16 @@ Do deep learning classification on the [CIFAR-10 database](https://www.cs.toront
 ### Keras Code Sample
 It's referenced the structure of [Keras code sample of CIFAR-10](https://keras.io/examples/cifar10_cnn/).
 <br>
-![](./model_summary/keras_sample_cifar10.png)<br>
-Test Accuracy : ≈78%<br>
+![](./model_summary/keras_sample_cifar10.png)
+
+- Test Accuracy : ≈78%
 ![](train_history/keras_sample_cifar10.png)
 
 ### Advanced Solution
 Next, let's referenced [the tutorial from Jason Brownlee PhD](https://machinelearningmastery.com/how-to-develop-a-cnn-from-scratch-for-cifar-10-photo-classification/), who is a professional developer and machine learning practitioner.<br>
-![](./model_summary/jason_brownlee_cifar10.png)<br>
-Test Accuracy : ≈89% (Train / Valid : 94.60% / 89.35%)<br>
+![](./model_summary/jason_brownlee_cifar10.png)
+
+- Test Accuracy : ≈89% (Train / Valid : 94.60% / 89.35%)
 ![](./train_history/jason_brownlee_cifar10.png)
 
 
@@ -21,8 +23,9 @@ Ref. [https://arxiv.org/abs/1412.6071](https://arxiv.org/abs/1412.6071)
 <br>
 Due to VGG-16 or ResNet-50 are so giant and deep neural network, I tried fractional max-pooling after read the research paper to make a deeper VGG-like neural network but smaller and  shallower than VGG-16 and ResNet-50. To reduce overfitting, we use global average pooling layer instead of full connection layer.
 <br>
-![](./model_summary/fmp_cifar10.png)<br>
-Test Accuracy : ≈93% (Train / Valid : 97.94% / 93.55%)<br>
+![](./model_summary/fmp_cifar10.png)
+
+- Test Accuracy : ≈93% (Train / Valid : 97.94% / 93.55%)
 ![](./train_history/fmp_cifar10.png)
 
 ### Simple Comparison