COURSE INFORMATION FOR THE CUGS COURSE IN ARTIFICIAL INTELLIGENCE
INFORMATION FOR COURSE PARTICIPANTS, VERSION 1.0
Goal for the course: The course is organized under the assumption
that your goal, as a participant, will be to acquire a broad and
systematic knowledge and understanding of modern artificial intelligence.
"Systematic" means that existing cross-connections within the topic
should be understood.
Remark: It is debatable to what extent Artificial Intelligence
has a common core and to what extent it is a loose collection of
separate techniques. In any case the course will try to elucidate
existing connections without pretending that they are more than
they are.
Assumed background: An undergraduate course in A.I. is assumed as
background. Concretely, I assume that as a participant you have
acquired selected parts of the following textbook:
Stuart Russell and Peter Norvig
Artificial Intelligence. A Modern Approach
Prentice-Hall
according to the reading list from this book for the IDA course
TDDA 58 (see webpage). Unless you have taken TDDA 58 (or TDDA 13)
prevously, you are strongly recommended to catch up on the prerequisites
during the months before the course.
Special orientation of the course. The TDDA 58/13 course which is
assumed background is in fact oriented towards 'crisp' artificial
intelligence, where the world is modelled in terms of discrete
objects having discrete-valued properties and entering in relations
that are assumed to be either true or false. Therefore, and also
because of the general focus of CUGS, the present course will try to
complement it by giving sufficient room for the following aspects
of the field:
- intelligent agent architectures that operate in continously
changing environments
- reasoning and acting under uncertainty
- learning
- perception and robotics
It will also of course fill in on aspects of crisp A.I. that
are not covered in the prerequisites.
Accomodating different backgrounds. Since the participants in
the course can be assumed to have different background knowledge,
we must arrange it in as flexible a way as possible. We shall
therefore use the Russell-Norvig book as the main literature
for the CUGS core course as well. I expect that at the conclusion of
the course, all participants will master the entire textbook material,
including both the prerequisites and the material in the present
course per se.
However, the material in chapters 6 through 10 will not be
addressed in the present course, or only very marginally so.
It belongs instead to the CUGS core course in knowledge
representation. Those limited parts of chapters 6-10 that have
already been covered in the prerequisites will of course be used
in the present course, but no new material will be added.
Note however that the KR course is likely to contain a lot of
other material as well.
An attached table indicates, for each of the chapters in the book,
whether it is considered as prerequisites or is covered in the
present course, or in the KR course. Note however that the table
omits some detail with respect to sections within chapters.
Example agent worlds. A course of this kind should be combined
with some laboratory exercises; only reading a book does not
give full understanding of the topic. The textbook defines some
simple, discrete-valued model worlds for exercises, and via a
webpage it provides a program library to be used there. Working
with this material for exercises is an important part of taking
the course.
The program library is written in Lisp, so please make sure that
your knowledge of this language is up to date when the course
starts, and that a Lisp system is installed in the computer system
you are using. See the book's webpage for details about how to
obtain a system.
Although the book's example worlds are fine for their purpose,
they only illustrate the 'crisp' view of A.I. In line with the
orientation of the course I am going to complement them with a
model world involving continous change, namely, a simple road
traffic world, called CARSIM. Software for CARSIM will be made
available for the course.
Webpage sources:
The webpage for the present course will be found under the
following page
IDA course TDDA 58/13
http://www.ida.liu.se/education/ugrad/courses/tf/TDDA58/
Russell & Norvig textbook webpage
http://www.cs.berkeley.edu/~russell/aima.html
Course segments. The course will be organized in four *segments*,
with the following contents. The percentages indicate rough ideas
of what part of the total work for the course may be involved.
Note that 3 credits means 15 full working days, so 7% ought to
be roughly one full day's work.
Segment 1 (10%). Setting the stage.
Introduction to architectural issues in intellilgent
robotic agents
Introduction to the Wumpus world and other example worlds in
the textbook, and associated software
Introduction to the CARSIM example world, and associated
software
Brief orientation about some research projects in the
area
Prerequisite text: chapter 2; chapter 6 sections 6.1, 6.2, 6.5
Primary text: Course notes
Segment 2 (20%). Uncertainty, probabilistic reasoning, and
decision-making
Primary text: chapters 14 - 17 (110 pages)
Segment 3 (35%). Planning, acting, and robotics
Prerequisite text: chapter 3 - 5 (search and game playing)
and 11 (planning). A discussion about these chapters will
be part of the course
Primary text: chapters 12, 13, 25 (85 pages)
An implementation experiment, somewhat bigger than for the
other segments
Segment 4 (35%). Learning, neural networks, and perception
Prerequisite text: chapter 18
Primary text: chapters 19, 20, 21, 24 (175 pages)
Course meetings. This is something I wish to discuss with both you
(the course participants) and your advisors. The following are two
possible arrangements:
Organization A (very extensive). One major meeting at the beginning
of the course, another meeting towards the end of the course. Four
hours for each meeting. Advising by e-mail, telephone, and
appointment between these two sessions.
Organization B (moderately intensive). The four segments are
covered mostly in sequence, with one meeting each time a segment ends
and another segment starts, as well as meetings at the beginning and
the end of the course. Course meetings can be co-located with the
meetings for another course during the same period. Between the
meetings, e-mail discussions in a mailgroup for the course,
besides binary communication (participant and course leader).
The following would be a plausible allocation of time.
Week# Meeting# Topic Hours
1 1 Introduction to segment 1 3
Introduction to sgm 2 2
General issues 1
3 2 Discussion of segment 1 1
Discussion of sgm 2 3
Introduction to sgm 3 2
6 3 Discussion of sgm 1 + project 1
Discussion of sgm 3 3
Introduction to sgm 4 2
8 4 Discussion of sgm 1 + project 1
Discussion of sgm 4 3
Exam 2
This makes a total of 24 hours in course meetings, including the exam.
Time budget. As a participant in the course, you are expected to
acquire the course knowledge by individual study to the largest
possible extent. Different participants are likely to have different
restrictions on your time. However, in order to make a reasonable estimate
of the workload for the course that both participants and course leader
can commit to, it is useful to have a 'time budget' for the course.
Like any other budget, it can be used for follow-up: I appreciate
if particpants give me feedback during and after the course as to
its degree of realism.
The following is then how I imagine a realistic way of allocating
time for the course, with days of full-time work as the unit.
Participation in course meetings: 4 days (not full days)
Between meetings 1 and 2:
Read up on segment 1 1 day
Read up on segment 2 1 day
Computing exercises 1 day
Between meetings 2 and 3:
Read up on segment 3 <1 day
Computing exercises 3 days
Between meetings 3 and 4:
Read up on segment 4 <2 days
Computing exercises 2 days
Cursive reading: chapters 26,27
and webpage materials <<1 day
This is a total of 15 days, which corresponds to the 3 credits
for the course.
The reading time has then been estimated as follows. Reading straight
through should take (generously) 2 minutes per page, which means
3.5 hours for reading 100 pages. Double this time in order to give
adequate chance of going into detail with those limited parts of
the text that require repeated reading. This means 8 hours, 6 hours,
and 12 hours for the textbook reading in segments 2, 3, and 4,
respectively. Allowing 1, 1, and 2 days gives some extra margin.
Continuations. If you like this course, as I hope, then there will be a
number of additional CUGS courses should also interest you, in particular:
- Core course in knowledge representation
- Soft computing
- Distributed A I
- Advanced course in knowledge representation