Neural Networks

CC524

Dr. ‘Mohamed Waleed’ Fakhr

waleedf@aast.edu

Fall 2020
Biological Neural Networks (Brain)
‫ القشرة الدماغية‬Cerebral Cortex
What a biological neural network look like
Synaptic weights between neurons
 Main motivation for this course
• To develop artificial neural networks
that mimic the nature and functions of
biological neural networks.
• Doing so, we would create machines
that learn how to recognize objects,
predict motions, plan actions, etc.
(Machine Learning)
 The 3 main neural network functions

• Classification: to recognize faces, objects,

scenes, speech, speakers, emotions, etc. Also,
to group similar objects (unsupervised learning)
• Prediction: to predict people motion, weather,
prices, etc.
• Planning: to be able to plan actions, correct and
adjust steps, (drive a car in a busy road, play
game of chess, etc. (Reinforcement Learning)
 Course Requirements
• 7th: 20 on written exam, 5 on lab exam and 5 on
project data collection and SW installation
• 12th: 15 on written exam, 5 on lab exam.
• Pre-final: 10 on project
• Final: 40
• The official book by Simon Haykin: Neural Networks,
A comprehensive foundation. Many online courses
in neural networks and deep learningn like this
excellent book:
http://neuralnetworksanddeeplearning.com/chap1.html

Tutorial work will be based on Python (Scikit-learn,
Tensorflow, Keras frameworks)

Project: Github and Kaggle, Colab
Expression and Emotion Recognition
Speech Recognition using Deep Neural
Networks
Image segmentation and
object detection
Time Series Prediction
Reinforcement learning for self driving cars
Speech2Face using GAN
 Emotion synthesis GAN

https://arxiv.org/pdf/1905.05416.pdf
Deep Neural Networks for Medical Image Classification
More Applications of Neural Networks
Handwritten digit/letter recognition
OCR of printed documents
Biometrics: voice, iris, finger print, face,
and gait recognition
Medical Image segmentation and
diagnosis
Network traffic modeling, crowd sensing
Text classification and topic identification
Sentiment analysis
Biomedical signal classification: ECG,
EEG, heart sound, etc.
 Course contents
1- Introduction, Biological Neural Networks and the Neuron
model.
2- The perceptron as a neuron model. Linear and nonlinear
decision boundaries.
3- Multilayer perceptron (MLP) and The Back-propagation
learning algorithm.
4- Deep neural networks (a): (2007-2015): deep auto-
encoder and layer-wise pre-training
5- Deep neural networks (b): The revolution (2015-Now):
deep CNN.
6- CNN applications (classification, segmentation, etc.)
7- Generative Adversarial Network (GAN)
8- Traditional Neural Networks
Difference # 1: Brains are analogue; computers are digital
Difference # 2: The brain uses content-addressable memory
Difference # 3: The brain is a massively parallel machine; computers are not
Difference # 4: Processing speed is not fixed in the brain; there is no system clock
Difference # 5 :Short-term memory is not like RAM
Difference # 6: Unlike computers, processing and memory are performed by the same
components in the brain
And many more
In 2006, Deep Neural Networks have gained huge interest and
importance again.

Since 2015, with algorithms and hardware advancements,

and with some new clever tricks, the field of deep neural
networks is at a revolution