This document provides an agenda and overview for a deep learning course. The agenda includes an introduction to program and course learning outcomes, the syllabus, class management tools, and an introduction to week 1 of deep learning. The syllabus outlines 15 weekly topics on deep learning concepts and algorithms. Example student projects are provided showing applications of deep learning to areas like computer vision, natural language processing, and games. The introduction to week 1 discusses artificial intelligence, machine learning, and deep learning definitions and provides an overview of programming assignments and deep learning in action.

1. Faculty Computing
Deep Learning
Week 1
Dr. Hasanul Fahmi, M.Kom

2. Agenda
• Program Learning Outcome
• Course Learning Outcome
• Syllabus
• Class Manager and E-Campus
• Deep Learning Week-1

3. Program Learning Outcome
Bachelor of Informatics
• PLO-1 Able to analyze complex problems in the field of informatics and apply principles of informatics and
other relevant disciplines to identify their solutions by taking into account insights from the advancements of
trans-disciplinary fields.
• PLO-2 Able to design, implement, and evaluate computing-based solutions that meet the computing needs of
a disciplinary program.
• PLO-3 Able to communicate proficiently in various professional contexts.
• PLO-4 Able to comprehend professional responsibilities and to conduct assessment based on appropriate
information in computing practices and legal and ethical principles.
• PLO-5 Able to effectively conduct the role of team leader or member in activities that are in accordance with the
discipline of the study program.
• PLO-6 Able to apply computer science theories and the basis of software development to develop computing-
based solutions.
• PLO-7 Able to analyze, design, and develop a Startup Business supported by information technology.
• PLO-8 Able to comprehend the basics of research and scientific writing in the field of informatics.

4. Course
Learning Out
come
Able to analyze complex
problems, identify problems and
get solutions to problems from
Deep Learning.
Able to design, implement, and
evaluate Deep Learning
Able to apply Deep Learning
theories to develop computing-
based solutions.

5. Class Mark
• Attendance = 5%
• Class Exercises & Participations = 15 %
• Assignments = 15%
• Projects = 15%
• Midterm = 20%
• Final Exam = 30%
Student with absent more than 4 may
not be allowed to sit for Final Exam.

6. Cours Syllabus
• Week 1- Intoduction
• Week 2- Deep Learning Intuition
• Week 3- Shallow Neural Network
• Week 4- Deep Neural Networks
• Week 5- Practical aspects of deep learning
• week 6- Optimization algorithms
• Week 7- Proposal Project Meetings
• Week 8 - Midterm
• Week 9 - Structuring Machine Learning Projects
• Week 10 - Convolutional Neural Networks
• Week 11 - interpretability of Neural Networks
• Week 12 - RRN Recurrent Neural Networks
• Week 13 - Deep Reinforcement Learning
• Week 14 - Natural Language Processing and Word
Embeddings
• Week 15 - Project Presentation
• Week 16 - Final Examination

7. An introduction
to Deep
Learning

8. What to
expect
• Artificial Intelligence, Machine Learning, Deep
Learning
• The 5 myths of AI
• Over View Programming Assigments
• Example of students projects
• Deep Learning in action
• Basics of Deep Learning
• NVIDIA Volta V100 and AWS P3
• Q&A

9. • Artificial Intelligence: design software applications which
exhibit human-like behavior, e.g. speech, natural language
processing, reasoning or intuition
• Machine Learning: teach machines to learn without
being explicitly programmed
• Deep Learning: using neural networks, teach machines to
learn from data where features cannot be explicitly
expressed

10. The 5 Myths of AI

11. Myth #1 - AI is
the flavour of
the month

12. Fact #1 - AI is 60 years
old
John McCarthy (1927-2011)
1956 - Coined the term “Artificial Intelligence”
1958 - Invented LISP
1971 - Received the Turing Award
Marvin Minsky (1927-2016)
1959 - Co-founded the MIT AI Lab
1968 - Advised Kubrick on “2001: A Space Odyssey”
1969 - Received the Turing Award

13. Myth #2 - AI is
dark magic
aka « You’re not
smart enough »

14. Fact #2 - AI is
math, code and
chips
A bit of Science, a
lot of Engineering

15. Myth #3 – The
“cognitive”
unicorn

16. Myth #3 – The
“cognitive” unicorn

17. Fact #3: AI is a wide range of techniques and tools
• Machine Learning
• Natural Language Processing
• Speech
• Vision
• Expert Systems
• And more

18. Myth #4 - AI
is for esoteric
use cases

19. Fact #4: AI shines
on intuitive
problems

20. Myth #5 - AI is not production-ready

21. y = 0 y = 1 y = 2 y = 3 y = 4 y = 5
1 
 
0
 
0
0
 
0
0
0
 
1 
 
0
0
 
0
0
0
 
0
 
1 
0
 
0
0
0
 
0
 
0
1 
 
0
0
0
 
0
 
0
0
 
1 
0
0
 
0
 
0
0
 
0
1 
Projects: SIGN language image classification

22. Kian Katanforoosh
Projects: others
Face recognition
85% Jeff
And many more…
Car detection
Music generation Text generation
Trigger word detection
Machine translation
Optimal goalkeeper shoot
prediction
“I love you”
Emojifier
Art generation

23. [Deep Learning Specialization]
Assignment: Car detection for autonomous driving

24. Projects: others
Face recognition
85% Jeff
And many more…
Car detection
Music generation Text generation
Trigger word detection
Machine translation
Optimal goalkeeper shoot
prediction
“I love you”
Emojifier
Art generation

25. [L. Gatys et al.: Image Style Transfer Using Convolutional Neural Networks , 2015]

26. [L. Gatys et al.: Image Style Transfer Using Convolutional Neural Networks , 2015]

27. [L. Gatys et al.: Image Style Transfer Using Convolutional Neural Networks , 2015]

28. Projects: others
LeafNet: A Deep Learning Solution to Tree Species Identification
[Galbally, Rao & Pacalin: Spring 2018,
http://cs230.stanford.edu/projects_spring_2018/posters/8285741.pdf] [Steven Chen: Fall 2017]
Predicting price of an object from a picture
Neural
Network
300$

29. Projects: others
[Culberg: Winter 2019,
Detect cards from real-time video of tournaments to improve
viewer understanding and accessibility

30. Projects: others
[Fernandez: Fall 2019,
font-gen: Deep Models for Inferring Alternate Language Sets
from Fonts

31. Projects: others
[Pontius, Sakata and Santos: Spring 2019,
NBA 2k19 DeepBaller: A NN-Controlled Real-Time video game AI

32. Projects: others
Image-to-Image translation with Conditional-GAN
[Hu, Yu & Yu, Spring 2018:

33. Projects: others
[Albrecht, Wang & Zhu: Spring 2019,
Discrete reasoning in natural language processing

34. AI+X: Leveraging your
subject-matter
expertise
• Roy, Quill, and Tuchman from Material Science & Engineering predicted the
melting point and viscosity of ionic liquids based on the component anion and
cation chemical structures (report poster).
Buechler from Mechanical Engineering developed a deep learning framework
to approximate the outputs from a power flow simulation, and evaluate
performance for a variety of power network characteristics (report poster).
Sokol and Aguirre from the Biomedical Informatics Training Program
leveraged deep learning to estimate the ancestral composition of a genomic
sequence at high resolution (report poster).
Peng, Zhao, Yu from Computer Science, Civil Engineering, and Biomedical
Engineering used deep learning to classify gestures from divers
communicating with an autonomous robot companion in dangerous underwater
environments (report poster).
O’Day, Seagers, and Lee from Bioengineering and Mechanical Engineering
studied neural signals of patients with Parkinson’s disease while walking to
predict freezing behaviors (report poster).
•
•
•
•

35. And many more…
Predicting atom energy based on atomic-structure
• Visual Question Answering
Cancer/Parkinson/Alzheimer detection
Activity recognition in video
• Music genre classification / Music Compression
Accent transfer in a speech
• Generating images based on a given legend
• Detecting earthquake precursor signals
• …

36. Fact #5: AI
means
business

37. Deep Learning in action

38. Basics of Deep Learning

39. The neuron Activation functions
&
! xi ∗ wi = u
i ( )
”Multiply and Accumulate”
Source: Wikipedia

40. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
x =
x11, x12, …. x1I
x21, x22, …. x2I
… … …
xm1, xm2, …. xmI
I features
m samples
y =
y1
y2
…
ym
m labels,
2
N outputs
0,0,1,0,0,…,0
1,0,0,0,0,…,0
…
0,0,0,0,1,…,0
One-hot encoding
Neural networks

41. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
x =
… .x1I
… .x2I
x11,
x21,
…
xm1,
x12,
x22,
… …
xm2, …
I features
m samples
y =
y1
y2
…
ym
m labels,
2
N categories
0,0,1, …,0
1,0,0, … ,0
.xmI
0,0,
0,0
,
0,1,
…
0,0,0, … ,0
One-hot encoding
Number of correct predictions
Accuracy =
Total number of predictions
Neural networks

42. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Neural networks
Initially, the network will not predict correctly
f(X1) = Y’1
A loss function measures the difference between
the real label Y1 and the predicted label Y’1
error = loss(Y1, Y’1)
For a batch of samples:
𝑏𝑎𝑡𝑐0 𝑠3𝑧𝑒
!
3()
loss(Yi, Y’i) = batch error
The purpose of the training process is to
minimize error by gradually adjusting weights

43. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Training
Training data set
Trained
neural network
Hyper parameters
Backpropagation
Batch size
Learning rate
Number of epochs

44. Stochastic
Gradient
Descent (SGD)
Imagine you stand on top of a mountain with
skis strapped to your feet. You want to get down
to the valley as quickly as possible, but there is
fog and you can only see your immediate
surroundings. How can you get down the
mountain as quickly as possible? You look around
and identify the steepest path down, go down
that path for a bit, again look around and find
the new steepest path, go down that path, and
repeat—this is exactly what gradient descent
does.
Tim Dettmers
University of Lugano
2015
https://devblogs.nvidia.com/parallelforall/deep-learning-nutshell-history-training/
The « step size » is called
the learning rate
z=f(x,y)

45. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Validation
Validation data set Trained
neural network
Validation
accuracy
Prediction at
the end of
each epoch
Save the model at the end of each epoch

46. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Early stopping
Training accuracy
Accuracy
100%
Loss function
Epochs
Validation accuracy
Loss
Best checkpoint
OVERFITTING

47. © 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Deep Learning in practice
AWS Deep Learning AMI
MXNet
Torch
CTK
Keras
Theano
Caffe
TensorFlow
Amazon EC2
Anaconda
Intel MKL
CUDA+cuDNN Python 2+3
Caffe2
• One-click launch
• Single node or distributed
• CPU, GPU, FPGA
• NVIDIA & Intel libraries
• Anaconda Data Science Platform
• Python w/ AI/ML/DL libraries

48. Thank you!