DEEP LEARNING NEURAL

NETWORK
 Handwritten Sequence

• Human being effortlessly can recognize these digits

•If we attempt to write computer program to recognize the
digits, it is an Herculean task in the field of pattern recognition.
 Why Deep Learning
• Non-linearities needed to learn complex representations of
data.

• Deep learning algorithms attempt to learn data representation

by using a hierarchy of multiple layers.

• More layers and neurons approximate more complex functions

Performance Vs. Sample Size
 Deep Learning

• Deep learning models high-level abstractions by using a deep

network with multiple processing layers, composed of multiple
linear and non-linear transformation.

• The term deep learning begins to gain popularity after a paper

by Hinton on 2000

• Deep learning based models learn representations of

large-scale unlabeled data.
• It replaces handcrafted features with efficient algorithms
for feature learning and hierarchical feature extraction
 Deep Learning
• As we construct larger neural networks and train
them with more and more data, their performance
continues to increase.
• This is generally different to other machine learning
techniques that reach a plateau in performance.
• Deep learning involves training a hierarchy of
feature detectors in comparison to shallow learning
models which rely on hand-crafted feature
detectors.[6]
 Deep Learning
•The first layer learns primitive features, occur more often, like
an edge in an image or the tiniest unit of speech sound.
•Once that layer learns the features, they are fed to the next layer
to learn complex features, like a corner or a combination of
speech sounds.

•The process is repeated in successive layers until the system

can reliably recognize phonemes or objects.
•Drivers behind deep learning: huge amount of unstructured
complex data increases performance with the size of the neural
network.
 Neural Network
• A neural network is a massively parallel distributed
processor made up of simple processing units that
has a natural propensity for storing experiential
knowledge and making it available for use. [2]

• It resembles the brain in two respects

• 1. Knowledge is acquired by the network from its
environment through a learning process.

• 2. Interneuron connection strengths, known as

synaptic weights, are used to store the acquired
knowledge.
Nonlinear Model of a Neuron
Activation Function (Threshold)
Activation Function (Sigmoid)

The sigmoid function saturates when its argument is very positive or

very negative, meaning that the function becomes very flat and
insensitive to small changes in its input.
 Human Nervous System [2]

Stimulus
Response
Receptors Neural Net Effectors
 Human Brain[2]
• Receptors – convert stimuli from the human body or
the external environment into electrical impulses that
convey information to the neural net (brain)
• Neural Net – continually receives information,
perceives it, and makes appropriate decisions
• Effectors – convert electrical impulses generated by
the neural net into discernible responses as system
output
• Arrows pointing from left to right indicate the forward
transmission of information-bearing signals through
the system
• Arrows pointing from right to left signify the presence
of feedback in the system
 Feed forward Neural Networks
• Multilayer Perceptrons
• Deep Feedforward Networks
• A feedforward network defines a mapping y = f (x;
θ) and learns the value of the parameters θ that
result in the best function approximation.
• There are no feedback connections in which
outputs of the model are fed back into itself.
• When feedforward neural networks are extended
to include feedback connections, they are called
recurrent neural networks.[7]
Feed-forward Network
Feed Forward Neural Network
 Convolution Neural Network
• Do we really need all the edges of the network?
• Can some of these be shared?
 Local Receptive field

Each neuron in the first hidden layer will be connected to a

small region of the input neurons

Hidden neuron

•The region in the input image is called the local receptive field for
the hidden neuron.
• Each connection learns a weight.
 Hidden Layer

• Slide the local receptive field across the entire input image and
by sliding one pixel to the right (i.e., by one neuron), connect
to a second hidden neuron and so on building the first hidden
layer.
• For each local receptive field, there is a different hidden
neuron in the first hidden layer.
First Hidden layer

A different stride length is also used

 Learning an image:

Can represent a small region with fewer parameters

“beak” detector
 Same pattern appears in different places:
They can be compressed!
What about training a lot of such “small” detectors
and each detector must “move around”.

“upper-left
beak” detector

They can be compressed

to the same parameters.

“middle
beak”
detector
 A convolutional layer
A convolutional layer has a number of filters that does
convolutional operation.

Beak
detector

A filter
 Convolution
These are the network
parameters to be learned.

1 -1 -1
1 0 0 0 0 1 -1 1 -1 Filter
0 1 0 0 1 0 1
-1 -1 1
0 0 1 1 0 0
1 0 0 0 1 0 -1 1 -1
-1 1 -1 Filter
0 1 0 0 1 0
2
0 0 1 0 1 0 -1 1 -1

…
…
6x6
image Each filter detects a small
pattern (3 x 3).
 1 -1 -1
Convolution
-1 1 -1 Filter
-1 -1 1 1
stride=1

1 0 0 0 0 1 Dot
product
0 1 0 0 1 0 3 -1
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0

6x6
image
 Convolution
1 -1 -1
-1 1 -1 Filter
-1 -1 1 1
stride=1

1 0 0 0 0 1
0 1 0 0 1 0 3 -1 -3 -1
0 0 1 1 0 0
1 0 0 0 1 0 -3 1 0 -3
0 1 0 0 1 0
0 0 1 0 1 0 -3 -3 0 1

6x6 3 -2 -2 -1
image
Convolution 1 -1 -1
-1 1 -1 Filter
If stride=2 -1 -1 1 1

1 0 0 0 0 1
0 1 0 0 1 0 3 -3
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0

6x6
image
Convolution -1 1 -1
-1 1 -1 Filter 2
-1 1 -1
stride=1
Repeat this for each filter
1 0 0 0 0 1
3 -1 -3 -1
0 1 0 0 1 0 -1 -1 -1 -1
0 0 1 1 0 0
1 0 0 0 1 0 -3 1 0 -3
-1 -1 -2 1
0 1 0 0 1 0 Featur
0 0 1 0 1 0 - -3e Map
0
- -1 1-2 1
3
1
6x6 3 -2 -2 -1
- 0 -4 3
image
1
Two 4 x 4 images
Forming 2 x 4 x 4 matrix
 Color image: RGB 3 channels

11 - - - - - - 11 - -
1 1 -11 -1 1 -1 1 1 -1
- - 111 - 1 -1 11 - 1
-1 1 - -1 Filter 1 1 -1 1 1- -1 Filter
-
1
-1 - 11 -1 11 - 1
1
- -1 - -1 11 - - -1 2
1 1 11 -1 1 11
Color 1 1
image 1 0 0 0 0 1
1 0 0 0 0 1
0 11 00 00 01 00 1
0 1 0 0 1 0
0 00 11 01 00 10 0
0 0 1 1 0 0
1 00 00 10 11 00 0
1 0 0 0 1 0
0 11 00 00 01 10 0
0 1 0 0 1 0
0 00 11 00 01 10 0
0 0 1 0 1 0
0 0 1 0 1 0
Convolution v.s. Fully Connected

1 0 0 0 0 1 1 -1 -1 -1 1 -1
0 1 0 0 1 0 -1 1 -1 -1 1 -1
0 0 1 1 0 0 -1 -1 1 -1 1 -1

1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0

convolution
image
fewer parameters!
1 0 0 0 0 1
0 1 0 0 1 0
Fully- 0 0 1 1 0 0
1 0 0 0 1 0
conne
…
…

…
…
0 1 0 0 1 0
cted 0 0 1 0 1 0
 Convolution and Shallow NN

• Convolutional Neural Networks are very similar to regular

Neural Networks.

• CNN are made up of neurons that have learnable weights and

biases, updated using loss function.

• Each neuron receives inputs, performs a dot product and

follows with a non-linearity.

• The CNN has an activation function (e.g. sigmoid/Softmax) on

the last (fully-connected) layer and all steps of learning regular
Neural Networks still apply.
 So what does change?
• It makes the forward function more efficient to implement and
vastly reduce the amount of parameters in the network.

• Regular Neural Nets don’t scale well to full images.

• For example, an image of size, 200×200×3 (200 wide,

200 high, 3 color channels), would lead to a single fully-
connected neuron in a first hidden layer with 200×200×3
= 120,000 weights.

• Full connectivity is wasteful and the huge number of

parameters would quickly lead to overfitting.
 Convolutional Neural Network
• A Convolutional Neural Network (CNN) is comprised of one
or more convolutional layers (often with a subsampling step)
and then followed by one or more fully connected layers as in
a standard multilayer neural network.

• Local connections and tied weights followed by some form of

pooling results in translation invariant features.

• Another benefit of CNNs is that they are easier to train and

have many fewer parameters than fully connected networks
with the same number of hidden units.
 Why Pooling

• Subsampling pixels will not change the object

Makes the network invariant to small transformations,
distortions and translations in the input image

Subsampling

We can subsample the pixels to make image smaller

Fewer parameters to characterize the image
 Max Pooling

1 -1 -1 -1 1 -1
-1 1 -1 Filter 1 -1 1 -1 Filter
2
-1 -1 1 -1 1 -1
3 -1 -3 -1 -1 -1 -1 -1

-3 1 0 -3 -1 -1 -2 1

-3 -3 0 1 -1 -1 -2 1

3 -2 -2 -1 -1 0 -4 3
 Max Pooling
New image
1 0 0 0 0 1 but smaller
0 1 0 0 1 0 Conv
3 0
0 0 1 1 0 0 -1 1
1 0 0 0 1 0
0 1 0 0 1 0 Max 30 13
0 0 1 0 1 0 Poolin
g 2 x 2 image
6x6
image Each filter is
a channel
The whole CNN

cat dog
…… Convolution

Max Pooling
Can
Fully Connected repeat
Feedforward
network
Convolution many
times

Max Pooling

Flattened
The whole CNN

3 0
- 1 Convolution
1
30 13
Max Pooling
Can
A new image
repeat
Smaller than the Convolution many
times
original image
The number of Max Pooling

channels is the number

of filters
 Fully Connected Layer of CNN

• The Fully Connected layer is a traditional Multi Layer

Perceptron that uses a softmax activation function in the output
layer.

• The term “Fully Connected” implies that every neuron in the

previous layer is connected to every neuron on the next layer.

• The output from the convolutional and pooling layers

represent high-level features of the input image.

• The purpose of the Fully Connected layer is to use these

features for classifying the input image into various classes
based on the training dataset.
The whole CNN

cat dog
…… Convolution

Max Pooling

Fully Connected A new image

Feedforward
Convolution
network

Max Pooling

Flattened A new image

 Flattening 3

1
3 0
-1 1 3

30 1 -
3 Flattene Fully Connected
d 1
Feedforward network
1

3
A CNN compresses a fully connected
network in two ways:

• Reducing number of connections

• Shared weights on the edges
• Max pooling further reduces the complexity
 Deep Learning

• Deep learning has an inbuilt automatic multi stage feature

learning process that learns rich hierarchical representations (i.e.
features).

Low Mid Level High Output

Trainable (e.g.
Level Features Level
Classifier outdoor,
Features Features
indoor)
 LAYERS USED TO BUILD
CONVNETS
• A simple ConvNet is a sequence of layers.

• Every layer of a ConvNet transforms one volume of

activations to another through a differentiable function.

• Three main types of layers to build ConvNet

architectures: Convolutional Layer, Pooling Layer,
and Fully-Connected Layer.

• Stack these layers to form a full ConvNet architecture.

Activation Functions

• Sigmoid neurons saturate and kill gradients

• If initial weights are too large then most

neurons would saturate

• Like Sigmoid, tanh neurons saturate.

•Unlike Sigmoid, tanh neurons zero

centered and scaled sigmoid.
 Non Linearity (ReLU)

• ReLU is a non-linear operation

• ReLU is an element wise operation applied per pixel and

replaces all negative pixel values in the feature map by zero.
• Trains much faster due to linear, nonsaturating form.

🙂 Prevents the gradient vanishing problem

 Overfitting

http://wiki.bethanycrane.com/overfitting-of-
data

Learned hypothesis may fit the

training data very well, even
outliers (noise) but fail to
generalize to new examples
(test data)

https://www.neuraldesigner.com/images/learning/selection_error.
svg
 Regularization

Dropout
• Randomly drop units (along with their connections) during training
• Each unit retained with fixed probability p, independent of other units
• Hyper-parameter p to be chosen (tuned)
 L2 = weight decay

• Regularization term that penalizes big weights, added to the

objective

• Weight decay value determines how dominant regularization is

during gradient computation.

• Big weight decay coefficient 🡪 big penalty for big weights

Early-stopping
• Use validation error to decide when to stop training
• Stop when monitored quantity has not improved after n subsequent
epochs
• n is called patience
 Loss functions and output
Classification Regression

Training Rn x {class_1, ..., class_n} Rn x

examples Rm (one-hot encoding)
Linear
Soft-max (Identity) or
Output
[map Rn to a probability Sigmoid
Layer
distribution]

f(x)=
x
Cost
(loss) Cross-entropy Mean Squared
function Error

Mean Absolute
Error

List of loss functions

 How can we Train Deep Networks?

• Deep networks not performing better than shallow networks

using stochastic gradient descent by backpropagation.

• Different layers in deep network are learning at vastly different

speeds.

• When later layers in the network are learning well, early layers
often get stuck during training, learning almost nothing at all.

• There's an intrinsic instability associated to learning by

gradient descent in deep, many-layer neural networks resulting
stuck at during training either at the early or the later layers.
 Dataset Shift

• In real world problems, often train and test datasets have not
been generated by the same distribution.

• This phenomenon has an adversarial effect on the quality of a

machine learning model, called dataset shift or drift.
 Covariate Shift

• Covariate shift is the change in the distribution of the

covariates specifically, that is, the independent variables.
 Covariate shift

• Mathematically, covariate shift occurs if ptrain(X) ≠

ptest(X) where X is a feature.

Say, an algorithm learned some X to Y mapping, now the

distribution of X changes, so we might need to retrain the
learning algorithm by trying to align the distribution of X with
the distribution of Y.

•However, the notion of covariate shift can be extended beyond

the learning system as a whole, but to apply to its parts, such as a
sub-network or a layer.
 Stochastic gradient descent(SGD)

• Stochastic gradient descent (SGD) has proved an effective way

of training neural networks.

• SGD optimizes the parameters θ of the network, so as to

minimize the loss

•With SGD the training proceeds in steps and each step we

consider a mini batch of size m, m << N.
•The mini batch is used to approximate the gradient of the loss
function w.r.t. the parameters by computing
 Stochastic Gradient

• Stochastic gradient is simple and effective, but it requires

careful tuning of the model hyper-parameters.
• Specifically the learning rate and the initial values for the
model parameters.
• The training is complicated by the fact that the inputs to each
layer are affected by the parameters of all preceding layers .
• Small changes to the network parameters amplify as
the network becomes deeper.
• The change in the distributions of layers’ inputs creates
problem because the layers need to continuously adapt to the
new distribution.
• A network computing ℓ = F2(F1(u,Θ1), Θ2) where F1 and F2
are arbitrary transformations, and the parameters Θ1, Θ2 are to
be learned so as to minimize the loss ℓ.

• Learning Θ2 can be viewed as if the inputs x = F1(u, Θ1) are

fed into the sub-network ℓ = F2(x, Θ2).

• A gradient descent step

for batch size m and learning rate α is exactly equivalent to that
for a stand-alone network F2 with input x.

• Therefore for training more efficiently – must having the same

distribution between the training and test data – apply to
training the sub-network as well.
 Internal Covariate Shift
• Fixed distribution of inputs to a sub-network would have
positive consequences for the layers outside the subnetwork, as
well.
• The change in the distributions of internal nodes of a deep network,
in the course of training, called Internal Covariate Shift.
• Eliminating it offers a promise of faster training.
• A new mechanism, call Batch Normalization reduces internal
covariate shift, which accelerates the training of deep neural nets.
• The goal of Batch Normalization is to achieve a stable distribution of
activation values throughout training.
 Batch Normalization
• We normalize the input layer when feature have values
wide spread and it speed up learning.

• We do the same thing for the values in the hidden layers,

and get 10 times or more improvement in the training
speed.

• Batch normalization reduces the amount by what the

hidden unit values shift around (covariance shift).
 Batch Normalization
• The distribution of each layer’s inputs changes during
training, as the parameters of the previous layers change.

• In a neural network, batch normalization is achieved through a

normalization step that fixes the means and variances of each
layer's inputs. Zero means, unit variances.

• The network becomes more robust to different initialization

schemes and learning rates.
 Batch Normalization
• Batch normalization is a technique for training very deep
neural networks that standardizes the inputs to a layer for each
mini-batch.

• Standardizing the activations of the prior layer means that

assumptions the subsequent layer makes about the spread and
distribution of inputs during the weight update will not change,
at least not dramatically.

• This has the effect of stabilizing the learning process and

dramatically reducing the number of training epochs required
to train deep networks.
 Implementation

• During training the mean and standard deviation of each input

variable to a layer is calculated per mini-batch and used to
perform the standardization.
• Allow the layer to learn two new parameters, namely a new
mean and standard deviation, Beta and Gamma respectively.
• It allows the automatic scaling and shifting of the
standardized layer inputs.
• These parameters are learned along with the original model
parameters, as part of the training process and restore the
representation power of the network.
 Implementation

• Batch normalization may be used on the inputs to the layer

before or after the activation function in the previous layer.

• It may be more appropriate after the activation function if for

s-shaped functions like the hyperbolic tangent and logistic
function.

• It may be appropriate before the activation function for

activations that may result in non-Gaussian distributions like
the rectified linear activation function.
 Issues

• If the mean and standard deviations for each input feature are
calculated over the mini-batch instead then the batch size must
be sufficiently representative of the range of each variable.

• In a batch-normalized model, we have been able to achieve a

training speedup from higher learning rates, with no ill side
effects.

• Batch normalization can make training deep networks less

sensitive to the choice of weight initialization method.

• Section – 8.7.1 Batch Normalization, Deep Learning, 2016.

 Batch normalization - Summary

• Batch normalization allows each layer of a network to learn by

itself, independently of other layers.
• Use higher learning rates because batch normalization makes sure
that there’s no activation that’s gone really high or really low.
• It reduces overfitting because it has a slight regularization effects.

• Similar to dropout, it adds some noise to each hidden layer’s

activations.
• Therefore, if we use batch normalization, we will use less dropout
or no drop out which is a good thing because we are not going to
lose a lot of information.
 CNN REVIEW

• The motivation of using fully connected networks for

image analysis with following benefits:

• Fewer parameters (weights and biases)

• Invariant to object translation
• Capable of generalizing and learning features.

• Convolutional layers are formed by filters, feature

maps, and activation functions.
We can determine the number of output layers of a given
convolutional block using number of layers in the input is nᵢ,the
number of filters in that stage, f, the size of the stride, s, and the
pixel dimension of the image, p(assuming it is square)
 Layers and Features

• Pooling layers are used to reduce overfitting.

• Fully connected layers are used to mix spacial and channel

features together.

• Each of the filter layers corresponds to the image after a

feature map has been drawn across the image.

• which is how features are extracted.

 Weights

• It is important to know the number of input and output layers

as this determines the number of weights and biases that make
up the parameters of the neural network.

• The more parameters in the network, the more parameters

need to be trained which results in longer training time.

• Training time is very important for deep learning as it a

limiting factor unless you have access to powerful computing
resources such as a computing cluster.
(i) 250 weights on the convolutional filter and 10 bias terms.
(ii)13 × 13 × 10 = 1,690 output elements after the max-pooling
layer.
(iii)200 node fully connected layer, which results in a total of 1,
690 × 200 = 338, 000 weights and 200 bias terms in the fully
connected layer.
(iv)338,460 parameters to be trained in the network. Majority of
the trained parameters occur at the fully connected output layer.
 Layer wise Complexity

• Each CNN layer learns filters of increasing complexity.

• The first layers learn basic feature detection filters such as

edges and corners.

• The middle layers learn filters that detect parts of objects —

for faces, they might learn to respond to eyes and noses.

• The last layers have higher representations: they learn to

recognize full objects, in different shapes and positions.
Feature maps showing increasing resolution of features through
different convolutional layers of a neural network.
 Transposed Convolution

• Upsampling of a convolution step, it is called transposed

convolution or fractional striding.

• A typical convolutional layer with no padding, acting on an

image of size 5 × 5. After the convolution, we end up with a 3
× 3 image.
• A convolutional layer with a padding of 1. The original image
is 5 × 5, and the output image after the convolution is also 5 ×
5.
• A padding of 2, original image 3× 3, and the output image
after convolution 5X5
• In the development of a variational autoencoder, these are
implemented using an upsampling layer.
 Deep Neural Networks

• Classical CNNs do not perform well as the depth of the

network grows past a certain threshold.

• There is a maximum threshold for depth with the traditional

CNN model.
 The Residual Block NN

• The failure of the 56-layer CNN could for the optimization

function, parameter initialization, or vanishing/exploding
gradient problem.

• The problem of training very deep networks has been

alleviated with the introduction of a new neural network layer
— The Residual Block.
 Skip Connection and Identity mapping

• Identity mapping does not have any parameters and is there to add
the output from the previous layer to the layer ahead.
• However, x and F(x) will not have the same dimension, because a
convolution operation typically shrinks the spatial resolution of an
image.
• The identity mapping is multiplied by a linear projection W for the
input x and F(x) to be combined as input to the next layer.

•
Ws term can be implemented with 1×1 convolutions,
introducing additional parameters to the model.
 Residual Networks

• The main idea behind this network is the residual

block.

• The network can contain 100 layers or more.

•With this extra connection, gradients can travel backward
more easily.
•It becomes a flexible block that can expand the capacity of the
network, or simply transform into an identity function that
would not affect training.
 A residual network stacks residual
blocks sequentially

•The idea is to allow the network to become deeper without

increasing the training complexity.
 Reference

• Kaiming He, et al. in their 2015 paper titled “Deep

Residual Learning for Image Recognition” used batch
normalization after the convolutional layers in their very
deep model referred to as ResNet and achieve then
state-of-the-art results on the ImageNet dataset, a standard
photo classification task.