09.05.2023 Views

pdfcoffee

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Chapter 4

Figure 10: Accuracy for different models and optimizers

Understanding the power of deep learning

Another test we can run for better understanding the power of deep learning and

convnets is to reduce the size of the training set and observe the resulting decay in

performance. One way to do this is to split the training set of 50,000 examples into

two different sets:

1. The proper training set used for training our model will progressively reduce

its size through 5,900, 3,000, 1,800, 600, and 300 examples.

2. The validation set used to estimate how well our model has been trained will

consist of the remaining examples. Our test set is always fixed, and it consists

of 10,000 examples.

With this setup we compare the previously defined deep learning convnet against

the first example neural network defined in Chapter 1, Neural Network Foundations

with TensorFlow 2.0. As we can see in the following graph, our deep network always

outperforms the simple network when there is more data available. With 5,900

training examples, the deep learning net had an accuracy of 97.23% against an

accuracy of 94% for the simple net.

[ 121 ]

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!