The first step of creating and training a new convolutional neural network (ConvNet) is to define the network layers. This topic explains the details of ConvNet layers, and the order they appear in a ConvNet.
The architecture of a ConvNet can vary depending on the types and numbers of layers included. The types and number of layers included depends on the particular application or data. For example, if you have categorical responses, you must have a softmax layer and a classification layer, whereas if your response is continuous, you must have a regression layer at the end of the network. A smaller network with only one or two convolutional layers might be sufficient to learn on a small number of gray scale image data. On the other hand, for more complex data with millions of colored images, you might need a more complicated network with multiple convolutional and fully connected layers.
You can define the layers of a convolutional neural network in MATLAB® in an array format, for example,
layers = [imageInputLayer([28 28 1]) convolution2dLayer(5,20) reluLayer maxPooling2dLayer(2,'Stride',2) fullyConnectedLayer(10) softmaxLayer classificationLayer];
layers is an array of
an input for the training function
Image input layer defines the size of the input images of a
convolutional neural network and contains the raw pixel values of
the images. You can add an input layer using the
imageInputLayer function. Specify your
image size in the
inputSize argument. Size of an
image corresponds to the height, weight, and the number of color channels
of that image. For example, for a gray scale image, the number of
channels would be 1, and for a color image it is 3.
In this layer, you can also specify the network to perform data augmentation, such as random flipping or cropping of the data, or data transformation by subtracting the mean of the image from the pixel values in the training set. The purpose of augmentation and normalization transforms is to reduce overfitting , which might occur with especially larger networks.
Filters and Stride: A convolutional
layer consists of neurons that connect to subregions of the input
images or the outputs of the layer before it. A convolutional layer
learns the features localized by these regions while scanning through
an image. You can specify the size of these regions using the
argument in the call to
For each region, the
computes a dot product of the weights and the input, and then adds
a bias term. A set of weights that are applied to a region in the
image is called a filter. The filter moves along
the input image vertically and horizontally, repeating the same computation
for each region, i.e., convolving the input. The step size with which
it moves is called a stride. You can specify
this step size with the
Stride name-value pair
argument. These local regions that the neurons connect to might overlap
depending on the
The number of weights used for a filter is h*w*c,
where h is the height, and w is
the width of the filter size, and c is the number
of channels in the input (for example, if the input is a color image,
the number of color channels is 3). The number of filters determines
the number of channels in the output of a convolutional layer. Specify
the number of filters using the
in the call to
Feature Maps: As a filter moves along the input, it uses the same set of weights and bias for the convolution, forming a feature map. Hence, the number of feature maps a convolutional layer has is equal to the number of filters (number of channels). Each feature map has a different set of weights and a bias. So, the total number of parameters in a convolutional layer is ((h*w*c + 1)*Number of Filters), where 1 is for the bias.
Zero Padding: You can also
apply zero padding to input image borders vertically and horizontally
'Padding' name-value pair argument.
Padding is basically adding rows or columns of zeros to the borders
of an image input. It helps you control the output size of the layer
it is added to.
Output Size: The output height and width of a convolutional layer is (Input Size – Filter Size + 2*Padding)/Stride + 1. This value must be an integer for the whole image to be fully covered. If the combination of these parameters does not lead the image to be fully covered, the software by default ignores the remaining part of the image along the right and bottom edge in the convolution.
Number of Neurons: The product of the output height and width gives the total number of neurons in a feature map, say Map Size. The total number of neurons (output size) in a convolutional layer, then, is Map Size*Number of Filters.
For example, suppose that the input image is a 28-by-28-by-3 color image. For a convolutional layer with 16 filters, and a filter size of 8-by-8, the number of weights per filter is 8*8*3 = 192, and the total number of parameters in the layer is (192+1) * 16 = 3088. Assuming stride is 4 in each direction and there is no zero padding, the total number of neurons in each feature map is 6-by-6 ((28 – 8+0)/4 + 1 = 6). Then, the total number of neurons in the layer is 6*6*16 = 256.
Learning Parameters: You can
also adjust the learning rates and regularization parameters for this
layer using the related name-value pair arguments while defining the
convolutional layer. If you choose not to adjust them,
the global training parameters defined by
For details on global and layer training options, see Set Up Parameters and Train Convolutional Neural Network.
A convolutional neural network can consists of one or multiple convolutional layers. The number of convolutional layers depends on the amount and complexity of the data.
The results from the neurons of a convolutional layer usually pass through some form of nonlinearity. Neural Network Toolbox™ uses rectified linear units (ReLU) function for this purpose. Each convolutional layer can be followed by ReLU layer or a pooling layer.
A convolutional layer is usually followed by a nonlinear activation
function. In MATLAB, it is a rectified linear unit (ReLU) function,
specified by a ReLU layer. You can specify the ReLU layer using the
reluLayer function. It performs a threshold
operation to each element, where any input value less than zero is
set to zero, i.e.,
The ReLU layer does not change the size of its input.
This layer performs a channel-wise local response normalization.
It usually follows the ReLU activation layer. You can specify this
layer using the
This layer replaces each element with a normalized value it obtains
using the elements from a certain number of neighboring channels (elements
in the normalization window). That is, for each element in
trainNetwork computes a normalized
where K, α,
and β are the hyperparameters in the normalization,
and ss is the sum of squares of the elements in
the normalization window .
You must specify the size of normalization window using the
in the call to
You can also specify the hyperparameters using the
K name-value pair arguments.
Note that, the previous normalization formula is slightly different
than what is presented in .
You can obtain the equivalent formula by multiplying the
Max- and average-pooling layers follow the
convolutional layers for down-sampling, hence, reducing the number
of connections to the following layers (usually a fully-connected
layer). They do not perform any learning themselves, but reduce the
number of parameters to be learned in the following layers. They also
help reduce overfitting. You can specify these layers using the
Max-pooling layer returns the maximum values of rectangular
regions of its input. The size of the rectangular regions are determined
For example, if
poolSize is [2,3], the software
returns the maximum value of regions of height 2 and width 3.
Similarly, the average-pooling layer outputs the average values
of rectangular regions of its input. The size of the rectangular regions
is determined by the
poolSize in the call to
For example, if
poolSize is [2,3], the software
returns the average value of regions of height 2 and width 3. The
scan through the input horizontally and vertically in step sizes you
can specify using the
Stride argument in the
call to either function. If the
poolSize is smaller
than or equal to
Stride, then the pooling regions
do not overlap.
For nonoverlapping regions (
equal), if the input to the pooling layer is n-by-n,
and the pooling region size is h-by-h,
then the pooling layer down-samples the regions by h . That is, the output of a max- or average-pooling
layer for one channel of a convolutional layer is n/h-by-n/h.
For overlapping regions, the output of a pooling layer is (Input
Size – Pool Size + 2*Padding)/Stride +
A dropout layer randomly sets the layer's
input elements to zero with a given probability. You can specify the
dropout layer using the
Although the output of a dropout layer is equal to its input,
this operation corresponds to temporarily dropping a randomly chosen
unit and all of its connections from the network during training.
So, for each new input element,
selects a subset of neurons, forming a different layer architecture.
These architectures use common weights, but because the learning does
not depend on specific neurons and connections, the dropout layer
might help prevent overfitting , .
Similar to max- or average-pooling layers, no learning takes place
in this layer.
The convolutional (and down-sampling) layers are followed by
one or more fully connected layers. You can specify a fully connected
layer using the
As the name suggests, all neurons in a fully connected layer
connect to the neurons in the layer previous to it. This layer combines
all of the features (local information) learned by the previous layers
across the image to identify the larger patterns. For classification
problems, the last fully connected layer combines them to classify
the images. That is why, the
in the last fully connected layer is equal to the number of classes
in the target data. For regression problems, the output size must
be equal to the number of response variables.
You can also adjust the learning rate and the regularization
parameters for this layer using the related name-value pair arguments
while defining the fully connected layer. If you choose not to adjust
trainNetwork uses the global training parameters
trainingOptions function. For details
on global and layer training options, see Set Up Parameters and Train Convolutional Neural Network.
For classification problems, a softmax layer and then a classification
layer must follow the final fully connected layer. You can specify
these layers using the
classificationLayer functions, respectively.
The output unit activation function is the softmax function:
where and .
The softmax function is the output unit activation function after the last fully connected layer for multi-class classification problems:
where and . Moreover, , is the conditional probability of the sample given class r, and is the class prior probability.
The softmax function is also known as the normalized exponential and can be considered the multi-class generalization of the logistic sigmoid function .
A classification output layer must follow the softmax layer.
In the classification output layer,
the values from the softmax function and assigns each input to one
of the k mutually exclusive classes using the cross
entropy function for a 1-of-k coding scheme :
where is the indicator that the ith sample belongs to the jth class, is the parameter vector. is the output for sample i, which in this case, is the value from the softmax function. That is, it is the probability that the network associates ith input with class j, .
You can also use ConvNets for regression problems, where the
target (output) variable is continuous. In such cases, a regression
output layer must follow the final fully connected layer. You can
specify the regression layer using the
The default activation function for the regression layer is the mean
where is the target output, and is the network's prediction for the response variable corresponding to observation i.
 Murphy, K. P. Machine Learning: A Probabilistic Perspective. Cambridge, Massachusetts: The MIT Press, 2012.
 Krizhevsky, A., I. Sutskever, and G. E. Hinton. "ImageNet Classification with Deep Convolutional Neural Networks. " Advances in Neural Information Processing Systems. Vol 25, 2012.
 LeCun, Y., Boser, B., Denker, J.S., Henderson, D., Howard, R.E., Hubbard, W., Jackel, L.D., et al. ''Handwritten Digit Recognition with a Back-propagation Network.'' In Advances of Neural Information Processing Systems, 1990.
 LeCun, Y., L. Bottou, Y. Bengio, and P. Haffner. ''Gradient-based Learning Applied to Document Recognition.'' Proceedings of the IEEE. Vol 86, pp. 2278–2324, 1998.
 Nair, V. and G. E. Hinton. Rectified linear units improve restricted boltzmann machines. In Proc. 27th International Conference on Machine Learning, 2010.
 Nagi, J., F. Ducatelle, G. A. Di Caro, D. Ciresan, U. Meier, A. Giusti, F. Nagi, J. Schmidhuber, L. M. Gambardella. ''Max-Pooling Convolutional Neural Networks for Vision-based Hand Gesture Recognition''. IEEE International Conference on Signal and Image Processing Applications (ICSIPA2011), 2011.
 Srivastava, N., G. Hinton, A. Krizhevsky, I. Sutskever, R. Salakhutdinov. "Dropout: A Simple Way to Prevent Neural Networks from Overfitting." Journal of Machine Learning Research. Vol. 15, pp. 1929-1958, 2014.
 Bishop, C. M. Pattern Recognition and Machine Learning. Springer, New York, NY, 2006.