Convolutional Neural

Networks

Gaurav Mittal
2012CSB1013
IIT Ropar
Lenet-5 (Lecun-98), Convolutional Neural Network for digits recognition [email protected]

1
 ANN Recap

gasturbinespower.asmedigitalcollection.asme.org 2
 What are CNNs?

Essentially neural networks that use

convolution in place of general matrix
multiplication in at least one of their layers.

http://goodfeli.github.io/dlbook/contents/convnets.html 3
Motivation

4
 Detection or Classification Tasks

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 5
 What to do with this data?

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 6
 Feature Representations

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 7
 Feature Representations

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 8
How is computer perception done?

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 9
Feature Representations???

10
 Computer Vision Features

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 11
 Audio Features

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 12
 NLP Features

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 13
 Certainly, coming up with features is
difficult, time-consuming and requires
expert knowledge.

A lot of time is spend tuning the

features which are often hand-crafted!

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 14
 Feature Representations

www.cse.ust.hk/~leichen/courses/FYTG.../FYTGS5101-Guoyangxie.pdf 15
 Feature Representations

www.cse.ust.hk/~leichen/courses/FYTG.../FYTGS5101-Guoyangxie.pdf 16
 Learning non-linear functions

www.cse.ust.hk/~leichen/courses/FYTG.../FYTGS5101-Guoyangxie.pdf 17
 Learning non-linear functions

Shallow

Deep

www.cse.ust.hk/~leichen/courses/FYTG.../FYTGS5101-Guoyangxie.pdf 18
 Biologically Inspired!

Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning 19
Features Learned by Deep Training

20
21
22
23
 Distinguished Features

Locally Receptive Fields

Shared Weights

Spatial or Temporal Sub-sampling

http://neuralnetworksanddeeplearning.com/chap6.html 24
 Typical CNN Layer

Convolutional Detector Normalization

Output:
Input Stage: Affine Stage: Pooling Stage Stage
Feature Map
Transform Nonlinearity (Optional)

25
 Typical CNN Layer

Convolutional Detector Normalization

Output:
Input Stage: Affine Stage: Pooling Stage Stage
Feature Map
Transform Nonlinearity (Optional)

26
 Convolution
 The convolution of f and g, written as f∗g, is defined as the integral of
the product of the two functions after one is reversed and shifted:

 Convolution is commutative.
 Can be viewed as a weighted average operation at every moment (for
this w need to be a valid probability density function)
 Discrete Convolution (one-axis):

https://www.wikipedia.org/ 27
 Cross-Correlation
• For continuous functions f and g, the cross-correlation is defined as:

where f* denotes the complex conjugate of f and τ is the lag

• Again, cross-correlation is commutative
• For discrete functions, it is defined as:

https://www.wikipedia.org/ 28
Convolution and Cross-Correlation
in Images
For a 2-D image H and a 2-D kernel F,

29
 How do they differ?
 Convolution is equivalent to flipping the
filter in both dimensions (bottom to top,
right to left) and applying cross-
correlation

 For symmetric kernels, both result in the

same output.

 Many machine learning libraries

implement cross-correlation but call it
convolution!

30
 2-D Convolution (without kernel
flipping)

Example of 'valid' 2-D convolution

(without kernel flipping) where a
3x4 matrix convolved with a 2x2
kernel to output a 2x3 matrix

http://goodfeli.github.io/dlbook/contents/convnets.html 31
 2-D Convolution in Action!

http://i.stack.imgur.com/I7DBr.gif 32
 Variants
Full • Add zero-padding to the image enough for every pixel to be visited
k times in each direction, with output size: (m + k - 1) x (m + k - 1)

Valid • With no zero-padding, kernel is restricted to traverse only within

the image, with output size: (m - k + 1) x (m - k + 1)

Same • Add zero-padding to the image to have the output of the same size
as the image, i.e., m x m

Stride s
• Down-sampling the output of convolution by sampling only every s pixels in each direction.
m−k+s
• For instance, the output of 'valid' convolution with stride s results in an output of size x
s
m−k+s
s
http://goodfeli.github.io/dlbook/contents/convnets.html 33
Why Convolution?

34
Why Convolution?

35
Local Receptive Field/Sparse Connectivity
 Convolution exploits the property of spatial local-
correlations in the image by enforcing local connectivity
pattern between neurons of adjacent layers

 Drastic reduce in the number of free parameters

compared to fully connected network reducing
overfitting and more importantly, computational
complexity of the network.

36
 Indirect Global Connectivity
• Receptive fields of units in deeper layers larger
than shallow layers

• Though direct connections are very sparse,

deeper layers indirectly connected to most of
the input image

• Effect increases with strided convolution or

pooling

37
 Example

Input neurons representing a

28x28 image (such as from
MNIST dataset)

http://neuralnetworksanddeeplearning.com/chap6.html 38
 Example

Every hidden layer neuron has a

local receptive field of region
5x5 pixels

http://neuralnetworksanddeeplearning.com/chap6.html 39
 Example

And so on, the first hidden layer is built!

(28 - 5 + 1) = 24 x 24 neurons in the hidden layer on 'valid' convolution

Size of the hidden layer can be changed using another variant of convolution

http://neuralnetworksanddeeplearning.com/chap6.html 40
 Shared Weights and Bias
• All neuron in the hidden layer share the
same parameterization (weight vector
and bias) forming a 'Feature Map‘

• (Shared Weights, Bias) →Kernel/Filter

• Now, the gradient of a shared weight is

sum of the gradients of the parameters
being shared.

41
 Shared Weights and Bias
• Translation Equivariance
o Allows features to be detected regardless of their position in the visual field.
(Feature is a kind of input pattern that will cause a neuron to activate, for eg. an
edge)

o All neurons in the first hidden layer detect exactly the same feature, just at
different locations.

o CNNs are well adapted to translation invariance of images: move a picture of a cat,
and it's still an image of a cat!

• Further reduces the number of free parameters, achieving better

generalization and computational performance.

42
 Typical CNN Layer

Convolutional Detector Normalization

Output:
Input Stage: Affine Stage: Pooling Stage Stage
Feature Map
Transform Nonlinearity (Optional)

43
Non-Linear Activation Function
• Sigmoid:

• Tanh:

• Rectified Linear Unit (ReLU):

Sigmoid Tanh

Most popular activation function

for DNN as of 2015, avoids
saturation issues, makes learning
faster
ReLU
44
 Feature Map - Obtained by convolution of the image with a linear filter, adding a bias
term and applying a non-linear function

 Require a number of such feature maps at each layer to capture sufficient features in
the image

 Let 𝑘 𝑡𝑡 feature map at a given layer be 𝑥 𝑘 , whose filters are determined by 𝑊𝑘 and bias
𝑏𝑘 , then 𝑥 𝑘 with sigmoid, 𝜎 function for non-linearity and filter of size m x m is
obtained as:
𝑚−1 𝑚−1

𝑥 𝑘 𝑖𝑖 = 𝜎 𝑊 𝑘 ∗ 𝑎 ij + 𝑏𝑘 = 𝜎 � � 𝑤𝑎𝑎 𝑦 𝑘−1 𝑖+𝑎 𝑗+𝑏

+ 𝑏𝑘
𝑎=0 𝑏=0
𝑖𝑖

45
• Each hidden layer is compose of
multiple feature maps, 𝑥 𝑘 , 𝑘 = 0. . 𝐾

• Weights, W of a hidden layer can be

represented in a 4D tensor containing
elements for every combination of
destination feature map, source feature
map, source vertical position, and
source horizontal position.

• Biases, b can be represented as a vector

containing one element for every
destination feature map.

46
 Typical CNN Layer

Convolutional Detector Normalization

Output:
Input Stage: Affine Stage: Pooling Stage Stage
Feature Map
Transform Nonlinearity (Optional)

47
 Pooling
 Non-linear down-sampling to simplify the information in
output from convolutional layer.

 Variants:
 Max pooling (popular)
 Weighted average based on distance
 L2 norm of neighborhood

 Reduces computation for upper layers by reporting summary

statistics (only with stride > 1)

 Provides translation invariance (Infinitely strong prior that

learning must be invariant to small translations) Bottom view has been shifted by 1 pixel w.r.t.
Top view.
Every value in the bottom row has changed, but
 Useful property, if we care more about whether some feature is only half the values in the top row has changed!
present than exactly where it is, thus adds robustness to
position
48
 Typical CNN Layer

Convolutional Detector Normalization

Output:
Input Stage: Affine Stage: Pooling Stage Stage
Feature Map
Transform Nonlinearity (Optional)

49
 Normalization (Optional)
Locally the response is normalized using some distance based weighted
average function

50
 Putting It All Together!

Lenet-5 (Lecun-98), Convolutional Neural Network for digits recognition

Lecun 1998 51
 Backpropagation
• Loss function
o For Classification
• Softmax Function with negative log likelihood

o For Regression
• Mean squared error

• Weight Update

where 𝜂 - learning rate,

𝛼 - momentum,
𝜆 - weight decay
52
 Backpropagation
• Convolutional Layer
o With error function, E, and filter output 𝒙𝒍 ,

Thus, the error is propagated to the previous layer.

• Pooling Layer
o Do not actually learn themselves, just reduce the size of the problem by introducing sparseness.
o Reduces region of k x k size to a single value during forward propagation.
o Error propagated back to the place where it came from, thus errors are rather sparse. 53
http://andrew.gibiansky.com/blog/machine-learning/convolutional-neural-networks/
Theano

54
 What is Theano?
• Theano is a Python-based Math Expression Compiler whose syntax is
quite similar to NumPy.

• Open-source project developed and maintained by ML group at

Université de Montréal.

• User composes mathematical expressions in a high-level description

mimicking NumPy's syntax and semantics which allows Theano to
provide symbolic differentiation.

http://deeplearning.net/ 55
 Key Features
• Single implementation compatible
with both CPU and GPU.

• Theano on its own optimizes using

CUDA C++ for GPU.

• Easy to implement back-propagation

in CNN, as it automatically computes
all the mappings involved.

• Creates a graph with the various

inputs involved, differentiating
using chain rule. Fitting a multi-layer perceptron to simulated data
with SGD having 784 inputs, 500 hidden units, a
10-way classification and training 60 examples at
a time
http://deeplearning.net/ 56
 Sneak Peek into Theano...

http://deeplearning.net/ 57
Theano-based implementations for
Deep Learning
 Caffe
 Torch
 Keras

Other Frameworks:
 cuDNN
 DIGITS

58
Caffe

59
 Key Features
• Deep learning framework (essentially for training CNNs) developed by
Berkeley Vision and Learning Center (BVLC)

• Speed: Able to process over 60M images per day with a single Nvidia
K40 GPU, thus considered to be the fastest convnet implementation
available.

• Expressive Architecture: Allows models and optimization to be defined

as configuration files rather than hard-coding, with ability to switch
between CPU and GPU by a single flag.

www.caffe.berkeleyvision.org 60
 Sneak Peek into Caffe
Convolutional Layer Max Pooling Layer Solver

61
 Age and Gender
Classification using
Convolutional Neural
Networks
Gil Levi and Tal Hassner
The Open University of Israel
IEEE Workshop on Analysis and Modeling of Faces and Gestures (AMFG), at the IEEE
Conf. on Computer Vision and Pattern Recognition (CVPR), Boston, June 2015

62
 Overview

 Uses deep-convolutional neural networks (CNN) for the task of

automatic age and gender classification.

 Despite the very challenging nature of the images in the Adience

dataset and the simplicity of the network design used, the method
significantly outperforms existing state of the art by substantial
margins.

63
 Dataset - The Adience Benchmark
• Consists of images automatically uploaded
to Flickr from smartphones.

• Viewing conditions of these images are

highly unconstrained, thus capturing
extreme variations in head pose, lightning
conditions, blur, occlusion, expressions and
more.

• Includes roughly 26K images of 2,284

subjects. Faces from Adience benchmark (above) and
breakdown into different classes (below)

• For the tests, in-plane aligned version of the

faces is used.

64
 Network Architecture

All 3 RGB 96 filters 256 filters 384 filters Both fully connected Output to
channels size 3x7x7 size 96x5x5 size 256x3x3 layers contain 512 class labels
First, resized neurons followed by (age /
to 256 x 256, Each convolutional layer is followed by rectified ReLU and dropout layer gender)
then cropped linear operator (ReLU), max pooling layer of 3x3
to 227 x 227 regions with 2-pixel strides and a local
normalization layer
65
 Measures to reduce overfitting
• Lean network architecture using just 3 convolutional layers and 2 fully
connected layers considering the size of the dataset and labels involved (8 age
classes and 2 gender classes)

• Dropout learning: Randomly set the output value of network neurons to 0

with a dropout ratio of 0.5 (50% chance)

• Weight decay: Used to keep the magnitude of weights close to zero

• Data Augmentation: Took random crop of 227x227 from image of 256x256

and randomly mirrored it in each forward-backward training pass

All these measures help in keeping the number of free parameters in the
network low reducing complexity and thus over-fitting

66
 Experiments
5-fold cross validation based on pre-
specified subject exclusive folds
distribution

What they used

• Trained on Amazon GPU machine with
1,536 CUDA cores and 4 GB GDDR5 RAM

What I used
• Trained on Nvidia Quadro K2200 with
640 CUDA cores and 4 GB GDDR5 RAM
Solver
67
 Results
Gender Classification
Accuracy
Method
Paper Reimplementation
Single-Crop 85.9 ±1.4 86.7 ± 1.5
Over-Sample 86.8 ± 1.4 87.4 ± 0.9

Age Estimation
Accuracy
Method Paper Reimplementation
Exact One-off Exact One-off
Single-Crop 49.5 ± 4.4 84.6 ± 1.7 49.5 ± 3.6 85.4 ± 1.8
Over-Sample 50.7 ± 5.1 84.7 ± 2.2 50.6 ± 5.0 85.8 ± 1.5

68
 Results - Age Estimation Confusion
Matrix
Paper Reimplementation

Predicted Labels Predicted Labels

0-2 4-6 8-13 15-20 25-32 38-43 48-53 60- 0-2 4-6 8-13 15-20 25-32 38-43 48-53 60-

0-2 0.699 0.147 0.028 0.006 0.005 0.008 0.007 0.009 0-2 0.741 0.139 0 0.028 0 0 0 0.093

4-6 0.256 0.573 0.166 0.023 0.010 0.011 0.010 0.005 4-6 0.057 0.654 0.135 0.135 0 0 0 0.019
Actual Labels

Actual Labels
8-13 0.027 0.223 0.552 0.150 0.091 0.068 0.055 0.061 8-13 0 0.114 0 0.828 0.057 0 0 0

15-20 0.003 0.019 0.081 0.239 0.106 0.055 0.049 0.028 15-20 0.018 0.119 0.065 0.653 0.106 0.015 0.010 0.010

25-32 0.006 0.029 0.138 0.510 0.613 0.461 0.260 0.108 25-32 0.009 0.094 0.009 0.471 0.292 0.037 0.037 0.047

38-43 0.004 0.007 0.023 0.058 0.149 0.293 0.339 0.268 38-43 0.02 0 0 0.22 0.56 0.14 0.06 0

48-53 0.002 0.001 0.004 0.007 0.017 0.055 0.146 0.165 48-53 0 0.1 0.033 0.067 0.133 0.267 0.4 0

60- 0.001 0.001 0.008 0.007 0.009 0.050 0.134 0.357 60- 0.238 0.012 0 0.008 0 0 0 0.740

69
 References
• http://deeplearning.net/tutorial/
• http://goodfeli.github.io/dlbook/contents/convnets.html
• http://neuralnetworksanddeeplearning.com/chap6.html
• http://deeplearning.net/software/theano/tutorial/
• http://andrew.gibiansky.com/blog/machine-learning/convolutional-neural-networks/
• Andrew Ng: Deep Learning, Self-Taught Learning and Unsupervised Feature Learning
• www.caffe.berkeleyvision.org
• http://www.openu.ac.il/home/hassner/Adience/index.html
• https://www.wikipedia.org/
• www.cse.ust.hk/~leichen/courses/FYTG.../FYTGS5101-Guoyangxie.pdf

• LeCun, Yann, et al. "Gradient-based learning applied to document recognition."Proceedings of the IEEE 86.11
(1998): 2278-2324.
• Bergstra, James, et al. "Theano: a CPU and GPU math expression compiler."Proceedings of the Python for scientific
computing conference (SciPy). Vol. 4. 2010.
• Gil Levi and Tal Hassner, Age and Gender Classification using Convolutional Neural Networks, IEEE Workshop on
Analysis and Modeling of Faces and Gestures (AMFG), at the IEEE Conf. on Computer Vision and Pattern
Recognition (CVPR), Boston, June 2015

70