Transcript introduction

Intro to Artificial Intelligence
COURSE # CSC384H1F
Fall 2008
Sonya Allin
Note: many slides drawn from/inspired by Andrew Moore’s 1
lectures at CMU and Faheim Bacchus here at U of T.
First, a little bureaucracy ...
The course text book:
Artificial Intelligence: A Modern Approach
(2003), 2nd Edition, Stuart Russell and
Peter Norvig.
* recommended, but not required.
* Two copies are on 24 hour reserve in the
engineering library.
* Lecture notes cover much of the course
material and will be posted to the website
before class.
* The text is, however, a good resource to
have around the house.
2
Where you can get course info
...
The course web site:
http://www.cdf.toronto.edu/~csc384h/fall/
The course bulletin board:
https://csc.cdf.toronto.edu/bb/YaBB.pl?board=CSC384H1F
The course web site will be the primary source of information and announcements.
Updates to homeworks, clarifications, etc., will be posted here.
The bulletin board, by contrast, will be sporadically monitored.
If, then, you have a question: (1) check the website, (2) hit the bulletin boards, and, if all
else fails, (3) come to visit me or a TA.
3
Prerequisites for the class ....
1. Only the CGPA pre-requisite will be enforced. To waive
that, you have to go to the Undergraduate office.
2. A stats class in your past will make the class easier: STA
250 or STA 255/247/257, for example.
3. We will do some work in Prolog (a Prolog review will take
place during tutorial sessions).
4. Any required background material that you don’t know,
you will have to learn on your own.
4
How you will be graded ....
Course work:
* 4 Assignments (some programming, some Q&A).
* A Midterm.
* A Final.
Assignments will be worth 50% of your grade (12.5% each). The midterm will be worth
15% and the final worth 35%.
Late policy:
10% off for each late day. No assignments accepted more than 5 days late.
Please start your assignments early!! So you get a good sense as to how long they will
take early on.
Plagiarism:
Just don’t do it. See http://www.cs.toronto.edu/~fpitt/documents/plagiarism.html for the
meaning of plagiarism and the U of T policies about it.
5
Contact information & tutorials ....
Instructor:
Sonya Allin, [email protected]
TAs:
Jasper Snoek, [email protected]
Christian Fritz, [email protected]
Maryam Fazel-zarandi, [email protected]
Hani Safadi, [email protected]
My office hours will be generally be Friday afternoons, 3 to 5 p.m. in Bahen
2200. This first week, however, I will hold office hours on WEDNESDAY
between 3 and 5!
The TA’s office hours will be posted to the website this week.
6
Some important dates ....
Are on the course hand out.
7
Who is the AI community? How
do they define AI?
8
9
10
11
12
13
Why do they do AI research?
Old fashioned answer:
To make programs that behave like
the human brain behaves.
14
Why do they do AI research?
New fashioned answer:
To make programs that behave like
the human brain SHOULD behave.
15
What research questions do
they ask?
16
Some researchers in the middle
Are the theoreticians.
These are useful people to have at your cocktail party. They often
manage to bridge the divide between the red and blue people.
“Theoretical AI” questions:
* Do some ways of reasoning allow the world to be represented
more completely, more accurately, in more detail?
* Do some ways of reasoning allow more questions to be
answered? With less ambiguity? With more confidence?
* What problems are simply not solvable?
17
Some researchers in the middle
CAPTCHAs, for example, live somewhere in the
middle.
See Luis von Ahn, Manuel Blum, Nicholas Hopper, and John Langford.
CAPTCHA: Using Hard AI Problems for Security. In Eurocrypt.
18
Now, what is an “intelligent”
system?
19
In this class, we focus on
REASONING
specifically:
1. Knowledge representation (i.e. logics),
2. Reasoning (with logic and probability),
3. Decision making (i.e. searching and planning).
20
What we assume about
reasoning
To be amenable to computation, we are going to assume our
agents are RATIONAL and that their behavior can be expressed
quantitatively.
1. ideal behaviors must be able to be defined quantitatively,
and
2. how close a particular system’s behavior is to an ideal must be
measurable.
Quantitative characterizations of rationality have come from
diverse areas like logic (laws of thought), economics (utility
theory how best to act under uncertainty), and game theory (how
self-interested agents interact).
21
More details on our focus
We’ll be primarily focused on knowledge engineering ... not so
much on learning and populating structures with which to reason
based on data from real world.
For alternative approaches to AI, you may want to check out:
CSC321H:
“Introduction to Neural Networks and Machine Learning”
CSC411H
“Machine Learning and Data Mining”
22
Material the course covers
We will first look at various ways to search:
Uninformed Search
Heuristic Search
Game Tree Search
This material is covered in Chapters 3 to 6 of the textbook.
23
Material the course covers
Then we will look at ways to represent knowledge:
First Order Logics
This material is covered in Chapters 7 to 10 of the textbook.
Then we will look at ways to make plans using logic:
State Representation
Planning Graphs
Analyzing Reachability
This material is covered in Chapter 11 and 12 of the textbook.
24
Material the course covers
Finally, we will look at ways to both reason and make
plans in uncertain conditions, using some probability
Probabilistic Reasoning
Bayesian Networks
Expected Utilities
Probabilistic Planning
This material is covered in Chapters 13 to 16 of the textbook.
25
Other AI courses to check out
CSC321H:
“Introduction to Neural Networks and Machine Learning” CSC401H:
“Natural Language Computing”CSC411H
“Machine Learning and Data Mining”CSC412H:
“Uncertainty and Learning in Artificial Intelligence”CSC420H:
“Introduction to Image Understanding”CSC485H:
“Computational Linguistics”CSC486H:
“Knowledge Representation and Reasoning”
CSC487H:
“Computational Vision”
26
Questions at this point?
27