- [11/30] In class yesterday, we covered perceptrons and just
started learning about neural networks and backpropagation. Some of
the topics we covered on perceptrons were: the basic algorithm, and
details on choosing alpha (the learning rate) and a termination
condition. We also discussed why the perceptron learning rule works ---
how it is doing stochastic gradient descent. We also discussed the
delta learning rule which uses an unthresholded perceptron. Oh, also
"batch" approaches which do the "real" gradient descent on an error
function over the entire training data set.
Chapters 2 and 3 in our text touch on these things, but not too
clearly. If you are looking for other reference material, you could
try Chapter 19 of the Russell and Norvig text or the chapter in
Mitchell's Machine Learning text. Both are on reserve at the library.
- [11/30] Monday's quiz will be mostly on Bayesian classifiers,
though there may be a bit of perceptrons on there. Bayesian
classifiers are also covered in Mitchell's Machine learning text in
case you are looking for a reference.
- [11/26] Assignment 7 is out and due on Monday December 3. The Assignment 7 web page is up (with
support code and 1 data set).
- [11/25] (As announced in class last week,) Tommorow's quiz will
be on probability and Bayes' rule which we have covered in the past
1.5 classes. It will not cover Bayes classifiers which we will finish
on Monday. A brief review of these topics can be found in sections
19.1 and 19.2 in our text. Any introductory text on probability and
statistics will also cover these topics; you could also look at
chapter 14 of Russell and Norvig's "Artificial Intelligence: a modern
approach" which is on reserve in the library.
- [11/20] A student pointed out that the example.txt file
to illustrate decision tree learning had some mistakes on page 2.
(For example "patrons" was selected as the top level attribute, but it
still appears as a candidate in remaining split information
calculations.) I cannot reproduce the error, so I don't know what
happened. I have put corrected files below in the handouts section.
(This was handed out on 11/8.)
- [11/8] Monday's quiz will be entirely on decision trees. I would
expect you to know basic concepts relevant to decision trees. I'd
expect you to know the formula for calculating information but not for
the "gain ratio" or "split information". I'd highly recommend doing
the first problem from Assignment 6 for practice.
- [11/8] As announced in class today, I am throwing out Quiz 9 from
this past Monday. I've deemed it not worth grading...
- [11/8] Today in class we finished up decision trees. After
reviewing the concepts of classification problems and measuring
information (for the "information gain" heuristic) and the basic
algorithm (called ID3), we discussed:
- a complete example (see the handout from today)
- overfitting
- connection between learning decision trees and search and
propositional logic
- dealing with continuous attributes
- dealing with missing attributes
- the "gain ratio," an alternative heuristic to "information gain"
- rule post pruning
This material is not covered in our text. Two references (which are
both on reserve in the library) are:
- For the basic stuff: Chapter 18 of Russell and Norvig's book
"Artificial Intelligence: a modern approach"
- For practical extensions and more advanced stuff: Chapter 3 of Mitchell's
"Machine Learning" book.
- [11/7] The support code for assignment 6 is out. See the assignment 6 information page.
- [11/1] Monday's quiz will generally be on First Order Logic (FOL)
and topics covered in class this week. This may include basic
concepts/syntax/etc of FOL, inference in FOL (particularly using the
resolution inference rule), transforming sentences into conjunctive
normal form, and (quite likely) a proof using resolution.
- [10/25] Monday's quiz will cover propositional logic. There will
be a proof on the quiz. You should understand the basic concepts and
terminology, (e.g. completeness, soundness, inference, entailment, and
so on) both intuitively and the formal definitions. Note that
slides on logic were handed out in class today, are available outside
my office door, and are also available in the handouts section below.
- [10/25] As I've said in class this week, I will be out of town at
a conference next week. On Monday, Zie Kone will be giving a lecture
and administering the quiz. On Thursday, Prof. Selmer Bringsjord from
the Cognitive Sciences Department will be giving a lecture on theorem
proving. I have asked him to provide some questions for the next
Monday's quiz!
- [10/25] As announced in class today, the tentative schedule for
the rest of the assignments is:
- Assignment 5: Out 10/25, due 11/5 (written)
- Assignment 6: Out 11/5, due 11/19? (programming decision tree learning)
- Assignment 7: Out 11/19? due 12/3? (programming)
- [10/15] I've added a new section below for links.
- [10/14] As announced in class on Thursday, Monday's quiz may
cover alpha-beta pruning with MINIMAX search although there will not
be a problem like Problem 3 from Assignment 4. It could also cover
EXPECTIMAX and the basics of iterative improvement algorithms and
constraint satisfaction problems. I suggested that students read
chapter 11 and look at Exercise 11.2 in the text.
- [10/12] The web tester and oracles for Problems 1 and 2 from
Assignment 4 are up! (So is the "web tester tester" and a "syntax
checker".) See the Assignment 4 page for details.
- [10/9] Today in class we finished game playing search. There was
one handout, the slides from today's class. Remember that Problem 3
of Assignment 4 is due on Thursday. On Thursday, we will start
Chapters 10 and 11 from the text (the last searching topics). The
oracles for problems 1 and 2 on Assignment 3 were finished last week
and are waiting to go up on the web. The testers for problems 1 and 2
should be up this week.
- [10/5] Monday's (well, Tuesday's) quiz next week will have at
least one A* problem. (This is why solutions to problem 1 on
assignment 3 were handed out in class on Thursday; copies are
available outside my office door if you didn't get one.) You will not
be expected to prove the optimality of A* but you may be expected to
know its properties (with respect to optimality, completeness, etc.)
It may also cover handling repeated states and minimax search (but not
with alpha-beta pruning). Make sure you are up on Chapters 7-9 and
12.
- [10/5] There are two mistakes on the Assignment 3 solutions for
Problem 1. In iteration 6, node G on the OPEN list has its path
updated but not its cost. The cost should be 9.4, not 10.3.
In iteration 5, the f value for node C on the
open list should be 9.4, not 9.4.0.
- [10/4] The Assignment 4
web page is up (with support code)!
- [9/30] Tomorrow's quiz will cover the blind search (DFS, BFS,
iterative deepening, and uniform cost) and heuristic search (greedy,
A*). You will not be asked to prove the optimality of A*.
- [9/28] Revised slides from yesterday's class and other material
are available in the handouts section and in
the (new) online course material section.
- [9/24] Assignment 3 deadlines have been extended as follows:
- Problem 1 (written A*) is still due on Thursday 9/27. The usual
late policy and automatic extension apply (as amended).
- Problems 2-4 and 6 (programming) are due at midnight Sunday night
9/30. There is no automatic extension of this deadline. The first
tier late deadline is midnight Monday night 10/1, and the second tier
late deadline is 10am Thursday 10/4.
- Problem 5 (a written description of your ep-heuristic
procedure) is due in class on Monday 10/1. There is no automatic
extension of this deadline. The first tier late deadline is 5pm
Monday 10/1. The second tier late deadline is 10am Thursday 10/4.
- [9/24] Today we went through an A* example, did an overview of
the optimality proof for A*, and discussed monotonicity and the
properties of A* with a monotonic heuristic. There were two handouts:
Assignment 1 solutions and Assignment 2 solutions. Finish reading
Chapter 9.
- [9/24] Zie Kone has added an additional office hour from 4-5 on
Wednesdays.
- [9/21] Yesterday in class, we went over the implementation of the
Missionaries and Cannibal problems (and breadth first search) in
Scheme. We also started heuristic search and went over Greedy search
and the A* algorithm (which we'll continue on Monday). There was one
handout: copies of the slides I used during class. Continue reading
Chapter 9 for Monday.
Monday's quiz could cover Chapter 7, Chapter 8 through section
8.4, and Chapter 9 sections 9.1 and the first parts of 9.2 (not more
than what we covered in class on Thursday). You should know the
difference between state space and a search tree, the difference
between states and nodes, what an admissible heuristic is, and why we
like the A* search.
- [9/20] The Assignment 3
information page is up.
- [9/18] Assignment 3 is out and is due Thursday September 27. Make sure
you have a CS account and kerberos password, since we expect to only
be using the web tester for this next assignment. The Assignment 3
information page will be up soon...
- [9/18] Yesterday we started search, covering state space versus
search trees, breadth first search, depth first search, completeness,
optimality, and space and time complexity. There were two handouts:
Assignment 3 and an implementation of the Missionaries and Cannibals
problem in Scheme. Start reading Chapter 9 for Thursday.
- [9/14] It appears that the final examination for this course has
been scheduled for Friday December 14 from 11:30am - 2:30pm. I'll
confirm this later.
- [9/14] Today we covered Chapters 5 and 6 from the text. There
was one handout, Scheme IV. We went over some topics from Scheme III
(map, apply, and lambda) and the topics from Scheme IV (printing,
begin, tail recursion, and helper functions). Monday's quiz will
have some Scheme on it (list recursion, map, apply, and lambda) and
topics from Chapters 4--6. In particular, there will certainly be
something on Genetic Programming and probably a problem from the line
labeling covered in Chapter 6.
Read Chapters 7 and 8 for Monday when we will begin "search."
- [9/13] Please note new submission instructions for Assignment 2
if you do not have a CS account. See the Assignment 2 Submission page.
- [9/12] A student pointed out a mistake in an example on the
Scheme III handout. In Section 11.1, the cond form in the
change-colors procedure should be a case form instead.
- [9/11] We had to change Zie Kone's office hours. His new office
hours are Wednesdays from 12-2 (or by appointment).
- [9/10] Today we talked about genetic programming from Chapter 4
of the text, covered recursion patterns, and went through some example
problems. Read Chapter 6 on computer vision for Thursday.
- [9/6] Today we covered perceptrons and neural networks from
Chapters 2 and 3 of the text. We did not get into the mathematical
details of backpropagation. We talked about list procedures,
conditionals, and the section on Scheme III on functional programming.
There were two handouts: Scheme III and Assignment 2.
- [9/6] Assignment 2 is out, due Thursday September 13. Electronic
turn-in details and other information will appear on
the Assignment 2 information page.
- [9/6] For late Assignment 1 papers (and papers turned in under
the automatic extension until midnight tonight if you signed up in
class), I'd prefer receiving hardcopy. You can drop it off at my
office (slip it under the door if I'm not there). However, Amos Eaton
building will supposedly be locked between 6 and 7pm although I don't
believe it. For this assignment, you can email your solution to our
TA, Zie Kone (konez@cs.rpi.edu). Future submission by email will be
at his discretion.
- [9/6] Quiz 1 is on Monday September 10. This will hopefully be a
15 minute quiz. Possible topics include Chapters 1-5 from the text
and Scheme. I made a few specific comments about possible quiz
questions at the end of class today. Quizzes are closed book and
closed notes.
- [9/4] I updated the links to the xscheme.el and
xscheme.elc files. You must have these files in place to use
MIT Scheme through gnu-emacs. (Of course, this only works on UNIX
systems.) See the "Getting started..." page off "The Scheme page" (below).
- [9/3] One of the links I wrote down in class for getting Scheme
was incorrect. Go to "the Scheme page" (below), then to the "getting
started..." page. The correct links are there.
- [8/31] Chapters 1-3 of the text are online in the
class reserves section of the library. On the class
reserves page, search under instructor "Huang" (evidently I'm the
only one) or under course name "Introduction to Artificial Intelligence".
- [8/30] For the next class, skim through Chapter 3 of
the text. We'll be covering the main ideas of this chapter in class.
- [8/30] Assignment 1 is out and due next Thursday. The Assignment 1 Page has information
about this assignment (errata, clarifications, etc.) I will be
updating this page regularly, so check back.
- [8/30] The bookstore is out of our textbook. They
ordered more on the 28th which they expect to come in on Tuesday or Wednesday.
In the meantime, I dropped off my copy at the library, and they
will be scanning Chapters 1-3 which will be available through their
class reserves system. This should probably be available before
Friday afternoon.
- [8/30] Today we covered "Reactive Agents" (mostly
following Chapter 2 of the text) and went over the Scheme II handout.
We'll continue with the Scheme II handout next week. There were three
handouts: Scheme II, Top 10 keybindings in Emacs/Edwin for MIT Scheme,
and Assignment 1.
- [8/27] Today we covered "What is AI?", went over the course
syllabus in detail, and started learning Scheme. There were two
handouts: the course syllabus (available in the handouts section
below) and the Scheme I handout.
For Thursday's class, read Chapters 1 and 2 from our text, read
the Scheme I handout, and figure out how you're going to run Scheme
(download an implementation to your computer or such --- see the
Scheme page listed below).