Deep Learning for Computer Vision

Defining the variables for a fully connected layer

Let's define a simple linear classifier or perceptron by explaining the variables such as weights and bias. The values of these variables will be learned during computation. These are also referred to as parameters of the model. The weight variables are initialized with normal random distribution with the shape of input size and number of classes. The input size is 784 here as the image is reshaped into a single vector. The number of classes is 10 which is equal to the number of digits in the dataset. The bias variable is also initialized with random normal distribution with the size equal to the number of classes. The weights and bias are defined as follows:

weights = tf.Variable(tf.random_normal([input_size ...