Course DIT822

Software Engineering
for AI Systems
H23 Course 2023

1
Course Intro

2
 AI – Artificial Intelligence - Intro
What is AI?

What is artificial? What is intelligence?

• Something made or produced by • the ability to learn or understand or
human beings rather than to deal with new or trying
occurring naturally, especially as a situations.
copy of something natural. • also : the skilled use of reason.

• There are reasons why some things

are artificial and substitute real
things.

See: A Collection of Definitions of Intelligence

3
http://www.vetta.org/documents/A-Collection-of-Definitions-of-Intelligence.pdf
 Are animals intelligent? If so, which ones?
• Raise your hand if you think
that:
1. monkeys are intelligent
2. dogs are intelligent
3. dolphins are intelligent
4. octopuses are intelligent
5. worms are intelligent

4
 Artificial Intelligence Definitions
Artificial intelligence (AI, also machine intelligence, MI) is intelligence displayed
by machines, in contrast with natural intelligence (NI)

Artificial general intelligence (AGI) is the intelligence of a machine that could

successfully perform any intellectual task that a human being can.

Narrow artificial intelligence (narrow AI) is a specific type of AI in which a

technology outperforms humans in some very narrowly defined task

5
Turing test of AI.
Player C, the interrogator,
is given the task of trying to determine
which player – A or B – is a computer
and which is a human.

6
EX Machina as a Turing Test

https://www.youtube.com/watch?v=Tu1ajhotzj0

You know it is a computer (robot), but does it feel

Or just pretend to feel?

It is intelligent or just behaves as intelligent?

7
In this course we are working with
narrow AI

More info: History of artificial intelligence

https://en.wikipedia.org/wiki/History_of_artificial_intelligence

8
 Applications of narrow AI
Music recommendations

Spotify's "Discover Weekly"

feature works with:

• Collaborative filtering
• Acoustic pattern analysis
• Playlist analysis with NLP

9
 Applications of narrow AI
Task-specific robots

• OrionStar’s AI-Powered 5G
Robotic Coffee Master
• Can make up to 1000 cups
of coffee per day
• Three times faster than human
• Based on
• 3000 hours of AI learning
• 30000 hours of robotic
arm testing and machine
vision training

Source: https://www.techeblog.com/5g-orionstar-robotic-coffee-master-robot/
10
Applications of narrow AI

Source: BBC, August 30, 2022

https://www.bbc.com/news/
world-europe-62717599 11
In this course we will
focus on

12
Machine Learning vs Traditional Programming

PROGRAM DATA OUTPUT DATA

OUTPUT New PROGRAM new

DATA (MODEL) OUTPUT

13
 Machine learning workflow (process)

Data managamenet ML Modeling ML Model operation

Sa Amersh et al: Software Engineering for Machine Learning: A Case Study, ICSE 2019
15
 What shall we study in this course?

System evolution
and advanced tools
Feature Engineering Modeling AI Systems Life Cycle
AI and ethics Data labeling Model training
Model evaluation
Sa Amersh et al: Software Engineering for Machine Learning: A Case Study, ICSE 2019
16
17
Organizational matters

18
Course Teachers

Daniel Razan Sushant Teodor Tayssir

Strüber Ghzouli Kumar Pandey Fredriksson Bouraffa

19
In memory of Prof. Ivica Crnkovic (1955-2022)

• Previous coordinator of this course

• One of the founders of the research direction
of AI Engineering
• Director of Chalmers AI Research Centre
(CHAIR)
• Remembered as a kind person who always
had an open ear for others and influenced
many people's lives for the better

20
Intake questionnaire

• Fill out this questionnaire:

https://forms.gle/xQ9qTvqN4aY2S3kMA
• You have 5 minutes

21
Learning outcomes
What will you learn here?

• Basic Machine Learning methods

• Supervised learning
• Unsupervised learning
• Deep neural networks
• Reinforcement learning

• Developing AI systems – AI systems life cycle

• Processing different types of data for AI/Machine Learning

• AI and ethics
• Trends and tools in Software Engineering for developing AI systems
22
Learning outcome - details
Lectures

• Lecture 1 0 Intro, 1 Linear regression

• Lecture 2 2 Multiple linear regression
• Lecture 3 3 Linear descent criteria, Polynomial regression
• Lecture 4 4 Classification, Logistic regression, One-vs-all ML
• Lecture 5 5 Unsupervised learning
• Lecture 6 6 Neural networks
• Lecture 7 7 Deep learning
• Lecture 8 8 Recurrent neural networks
• Lecture 9 9 ML engineering
• Lecture 10 10 Data management SE for ML+AI
• Lecture 11 11 Ethics and AI
• Lecture 12 12 Advances in ML and AI
 Learning outcome - labs
Labs follows the lectures
• Lab 0 Intro to Python
• Lab 1 intro to linear algebra, jupyter
• Lab 2 Linear regression, multiple linear regression
• Lab 3 Logistic regression
• Lab 4 Classification, Logistic regression
• Lab 5 Unsupervised learning
• Lab 6 Neural networks
• Lab 7 Deep learning
• Lab 8 consultation
• Lab 9 Feature engineering
• Lab 10 Data management
• Examples from the exam
How will we work?

• The course is intensive

• Eight weeks lectures and labs
• Interactions in groups
• Homework – continuations of labs
• Examen

• Check Canvas Course PM for the schedule

25
Lectures

• Lectures are given in-person

• Students are strongly encouraged to attend – the science shows that
students who regularly attend lectures achieve better outcomes
• The lectures are interactive
• There are breakout sessions for discussion and reflexion in smaller groups
• You can always ask a question via raising your hand

26
Labs
• Provide an overview/repetition of the lectures
• Describe the assignment (the homework) – typically by starting with a solution

• Assignments – most of the labs have assignments

• 4 assignments in total, based on 10 labs. So, each assignment covers 2-3 labs.
• Groups of students work together. Recommended group size: 4
• Find your group (there will be time at the end of this lecture)
• In lecture hall today
• After lecture: in Canvas chat
• An assignment can be submitted 2 times! After that the lab remains not approved.

• Some of the labs are adopted from Andrew Ng: Machine Learning MOOC at
Coursera course
• You can later take this course on-line and submit the solutions following the rules
from the course
27
Assignments

• Each assignment can give you max 1 - 10 points

• Less than 5 points requires resubmission of the assignment
• All assignments must be submitted to Canvas before the given deadline

• Upon completion of all labs, to get approved, the labs will be

individually discussed with the teachers

• A student must be able to explain the assignment and the details he/she
did. Otherwise not possible to receive a pass

28
Recommendation:
Do your assignments as you follow the lectures
Keep pace

L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 LAB exam

w35 w36 w37 w38 w39 w40 w41 w42

Lab Begin date Due date
Assignment 0 2023-08-28 2022-08-31 Group formation
Assignment 1 ~2023-08-29 2022-09-12 Software Setup, Lab 1: intro linear algebra+python
Based on labs 1+2 Lab 2, Linear Regression
Assignment 2 ~2023-09-15 2022-09-26 Lab 3 - Polynomial regression
Based on labs 3-5 Lab 4 - Classification, Logistic regression
Lab 5 Unsupervised learning
Assignment 3 ~2023-09-29 2022-10-10 Lab 6 Neural networks
Based on labs 6-8 Lab 7 Deep Learning
Lab 8 Recurrent Neural Networks
Assignment 4 ~2023-10-11 2022-10-17 Lab 9 ML - Feature engineering
Based on labs 9-10 Lab 10 Data Management
Python support

• We will use Python, the standard language for machine learning

• Many of you might not have a Python background
• Our support for you:
• An interactive tutorial tomorrow afternoon, lead by Razan
Please show up prepared (instructions in Canvas announcement)
• A contact person for Python questions – Razan

30
Generative AI Policy (ChatGPT etc.)

• You are allowed to use generative AI to complete assignments

• However, then you need to document your use of generative AI

• You need to be able to explain your solutions a discussion meeting

with the teachers
• Submitting AI-generated solutions without understanding them will not
lead to a pass

31
Examination
1. Assignments – 3 HEC - 4 assignments
• Two possible grades:
G – Pass (if at least 50% of all points of all assignments)
U – Fail (otherwise)
• Individual discussion with teachers with questions about assignments
2. Written Exam – 4,5 HEC
• Four possible grades:
5 (if at least 85% of exam points)
4 (if at least 70% of exam points)
3 (if at least 50% at exam points)
U (otherwise)
3. Total grade
Grade from written exam if you passed assignments and written exam
32
Group formation

• See assignment 0 on Canvas

• Groups should ideally have 4 students. We accept 3- or 5-student groups
• Until the end of Thursday (August 31), one e-mail must be sent by:
• Your group, if you found a (complete or incomplete) group
• You individually, if you haven't found a group
• Opportunities for finding a group:
• Today's lecture: break and time at the end
• Chat on Canvas

33
Student representatives

• We need at least 3 student student representatives, ideally gender-

balanced

• Any volunteers?

34
Good Luck!

• Now we start with the first lecture….

35