ECE553/653 Neural Networks

Introduction

1
 Basic Info
• Meet time: Tuesday/Thursday 2:00 pm – 3:15
pm
• Location: Dooly Memorial 101
• Instructor: Prof. Mingzhe Chen
• Office: Room 413, McArthur Engineering
Building
• Email: [email protected]
• Office hours: Tuesday/Thursday 1:00 pm –
2:00 pm

2
 About Me and Lab
• Faculty: Mingzhe Chen
• Research interest:
– Machine learning and artificial intelligence (AI) for
wireless networks,
– Distributed/Federated learning fundamentals,
– Virtual reality over wireless networks,
– Unmanned aerial vehicle over wireless networks,
– Semantic Communications
– Digital Network Twins

3
 About Me and Lab
• Current Projects:
– Digital Network Twins: Mapping Next Generation
Wireless into Digital Reality

4
 About Me and Lab
• Current Projects:
– AI Enabled Harmonious Wireless Coexistence for
3D Networks

5
 Basic Info
• Introduce yourself
– Name
– Major
– Why do you select this course?
– What do you want to learn from this course?

6
 Textbook and References
• Textbook
– Deep Learning by Ian Goodfellow, Yoshua Bengio,
and Aaron Courville, MIT Press, 2016
– Reinforcement Learning: An Introduction by Sutton
and Barto, MIT Press, 1998
– Machine Learning by Tom Mitchell, McGraw Hill,
1997
– A Course in Machine Learning by Hal Daume III
– Machine Learning Lecture Notes by Andrew Ng
– Machine Learning for Intelligent Systems by Kilian
Weinberger
7
 Grading
• Homework assignments - 30%
• Midterm - 20%
• Projects - 45%
– Supervised learning (report) (20%)
• Machine learning
• Neural networks
– Reinforcement learning (report)
• Attendance/Participation – 5%
– Pop quizzes

8
 Course Policies
• Class Attendance Policy
– Classroom participation and attendance
constitutes 5% of the final score. A random
number of in-class pop quizzes will be randomly
conducted throughout the semester, which also
serve as a way to take attendance. Missing more
than 3 pop quizzes (without excuse) will result in 0
participate score. Missing all pop quizzes (without
excuses) will result in a failure of this course.

9
 Course Policies
• Class Attendance Policy
– If at some point in the semester you cannot
physically attend class sessions due to illness,
injury, quarantine or isolation, or other approved
absence, you must contact the instructor.
Unexcused absences from the classroom may
affect your grade or lead to failing the course.

10
 Course Policies
• Academic Ethics
– Academic dishonesty in any form will not be
tolerated. The instructor of this course supports
the University of Miami Honor Code. Cheating,
plagiarism, or other forms of academic dishonesty
in this course is subject to the provisions of the
Honor Code.

11
 Course Policies
• Class Recordings
– Students are expressly prohibited from recording any part
of this course. Meetings of this course might be recorded
by the University. Any recordings will be available to
students registered for this class as they are intended to
supplement the classroom experience. Students are
expected to follow appropriate University policies and
maintain the security of passwords used to access
recorded lectures. Recordings may not be reproduced,
shared with those not in the class, or uploaded to other
online environments. If the instructor or a University of
Miami office plans any other uses for the recordings,
beyond this class, students identifiable in the recordings
will be notified to request consent prior to such use.

12
 Concepts
• Artificial intelligence (AI) vs. machine learning
(ML) vs. deep learning (DL)

AI
ML
DL

13
 What is Artificial Intelligence ?
• Boring textbook answer
– Intelligence demonstrated by machines –
Wikipedia

• What others say:

– The science and engineering of making computers
behave in ways that, until recently, we thought
required human intelligence. – Andrew Moore,
CMU

14
Examples of Artificial Intelligence

15
 Examples of Artificial Intelligence
• Programming
Problem Rules Code

Problem
Rules
Find a maximum Code
Sort the number
number
[1,3,5,4,2] [1,2,3,4,5]
16
 Why We Need ML?
• Machine learning is used to solve the
problems that we cannot write down the rules
– Image and speech identification
– Natural language processing

• Dog identification

17
 What is Machine Learning?
• A favorite

Study of algorithms that improve their performance

at some task with experience
– Tom Mitchell, CMU

18
An Example of Machine Learning

19
 What is Deep Learning?
• DL is a newer area of ML
– DL uses multi-layered artificial neural networks to
deliver high accuracy in tasks such as object
detection, speech recognition, language
translation

• The strength of DL
– DL can automatically learn, extract or translate the
features from datasets such as images, video or
text, without needing traditional hand-coded code
or rules
20
An Example of Deep Learning

21
 Machine Learning vs. Deep Learning

Machine Learning Deep Learning

Performs well on small Performs well on large
Data
to medium datasets datasets
Requires significant
Hardware Able to function on CPU
computing power e.g., GPU
Features need to be
Features Learns features automatically
manually identified
Training time Quick to train Computationally intensive

22
A Brief History of Machine Learning

23
A Brief History of Machine Learning
• 1950s
– Samuel’s checker player
– Selfridge’s Pandemonium
• 1960s:
– Neural networks: Perceptron
– Pattern recognition
– Learning in the limit theory
– Minsky and Papert prove limitations of Perceptron
• 1970s:
– Winston’s ARCH
– Symbolic concept induction
– Expert systems and the knowledge acquisition bottleneck
– Quinlan’s ID3
– Michalski’s soybean diagnosis
– Scientific discovery with BACON

24
A Brief History of Machine Learning
• 1950s
– Samuel’s checker player
– Selfridge’s Pandemonium
• 1960s:
– Neural networks: Perceptron
– Pattern recognition
– Learning in the limit theory
– Minsky and Papert prove
limitations of Perceptron
• 1970s:
– Winston’s ARCH
– Symbolic concept induction
– Expert systems and the
knowledge acquisition
bottleneck
– Quinlan’s ID3
– Michalski’s soybean diagnosis
– Scientific discovery with BACON 25
A Brief History of Machine Learning
• 1950s
– Samuel’s checker player
– Selfridge’s Pandemonium
• 1960s:
– Neural networks: Perceptron
– Pattern recognition
– Learning in the limit theory
– Minsky and Papert prove
limitations of Perceptron
• 1970s:
– Winston’s ARCH
– Symbolic concept induction
– Expert systems and the
knowledge acquisition
bottleneck
– Quinlan’s ID3
– Michalski’s soybean diagnosis
– Scientific discovery with BACON 26
A Brief History of Machine Learning
• 1980s:
– Advanced decision tree and rule learning
– Explanation-based Learning (EBL)
– Learning and planning and problem solving
– Utility problem
– Analogy
– Cognitive architectures
– Resurgence of neural networks (connectionism, backpropagation)
– Valiant’s PAC learning theory
– Focus on experimental methodology
• 1990s:
– Data mining
– Adaptive software agents and web applications
– Text learning
– Reinforcement learning (RL)
– Inductive logic programming (ILP)
– Ensembles: bagging, boosting, and stacking
– Bayes net learning

27
A Brief History of Machine Learning
• 2000s:
– Support vector machines & kernel methods
– Graphical models
– Statistical relational learning
– Transfer learning
– Sequence labeling
– Collective classification and structured outputs
– Computer Systems Applications (Compilers, Debugging, Graphics, Security)
– E-mail management
– Learning in robotics and vision
• 2010s:
– Deep learning systems
– Learning for big data
– Bayesian methods
– Multi-task & lifelong learning
– Applications to vision, speech, social networks, learning to read, etc.
– ???

28
 Neural Networks are taking over!
• Neural networks have become one of the
major thrust areas recently in various pattern
recognition, prediction, and analysis problems

• In many problems they have established the

state of the art
– Often exceeding previous benchmarks by large
margins

29
Breakthroughs with Neural Networks

30
Breakthroughs with Neural Networks

31
Image Segmentation & Recognition

32
Breakthroughs with Neural Networks

33
 Breakthroughs with Neural Networks

• Captions generated entirely by a neural

network
34
 Breakthroughs with Neural Networks

• https://www.theverge.com/tldr/2019/2/15/18226005/ai-generated-fake-people-
portraits-thispersondoesnotexist-styl 35
What We Will Cover in this Course?
• Supervised learning
– Linear regression
– Logistic regression
– Model ensembles
– K-nearest neighbor and decision tree

• Unsupervised learning
– Clustering and K-means

• Reinforcement learning
– Temporal difference learning
– Q learning
– Deep reinforcement learning
– Multi-agent reinforcement learning

36
What We Will Cover in this Course?
• Neural Networks
– Perceptron
– Feedforward neural networks
– Convolutional neural networks
• LeNet-5, AlexNet, ZFNet, VGGNet, ResNet
– Recurrent neural networks
• LSTM, GRU, ESNs
– Graph neural networks
– Generative models
• VAE, GAN

• Advanced deep learning

– federated learning, meta learning, transformer
37
 What is This Class About?
• Introduction to machine learning and neural
networks

• Goal:
– After finishing this class, you should be ready to
get started on your first deep learning research
project.
– Programing!

38
 What This Class is NOT?
• NOT the target audience:
– People looking to understand latest and greatest
cutting-edge research (e.g., ChatGPT, etc)

• Not the goal:

– Teaching a toolkit. “Intro to TensorFlow/PyTorch”

39