- [12/4] A new version of the a7code.com file is
available. It corrects a bug in updating the running reward averages.
This bug would have also affected the number of visits reported by
get-visits-element and get-action-state-visits.
- [12/3] The remaining web testers are up.
- [12/3] The web tester for Assignment 7, problem 1 is up.
- [12/3] New versions of the support code are out, though these are
minor updates, so you may not need these procedures (though they could
be useful)
The new version of the a7code.com file (version 1.0.2)
corrects a bug in the corresponding-max procedure. The new
version of the a7header.scm file (version 1.0.2) adds a
procedure get-reward-vector.
See the assignment 7 web page for details.
- [12/2] A new version of the a7code.com file (version
1.0.1) is available. It corrects a bug which called the
transform-state procedure while playing the dealer's hand
whereas it should not have.
- [12/1] A new version of the a7header.scm file (version 1.0.1) is
available. See the A7 page for details.
- [11/27] New versions of the assignment 7 support code are available.
- [11/27] Quiz 12 topics: Expect 1 or 2 questions from Quiz 11.
The main topics for Quiz 12 are utilities, sequential decision
problems, and reinforcement learning (what we covered in class on
Thursday 11/20 and Monday 11/24). The slides I showed were by the
authors of our text. Here are links:
I did not have prepared slides for our reinforcement learning.
- [11/24] Here are the Connect 4 tournament results
- [11/23] Note on Quiz 11: I will provide the formulas for the
Bayes classifiers. However, you will have to know which formula is
which. I expect you to know Bayes rule and the definition of
conditional probability.
- [11/23] Here is a Naive Bayes
classifier example.
- [11/23] Quiz 11 topics: Expect 1 or 2 questions from quiz 10.
(Solutions are on WebCT). The main topic is probability and bayesian
classifiers (brute force, optimal, and naive). There are slides on
WebCT that have the stuff we covered in class, although not all of it
in much detail. There is also some stuff on the slides (primarily
Bayesian updating) that we did not cover in class. You can refer to
chapter 13 for the basic probability stuff. The Bayesian classifiers
are covered in some detail in the slides.
Eric is working on a Bayes naive classifier example that we'll post
when he's finished with it.
- [11/17] Information on A6p4 and A6p5 is now on the Assignment 6
web page.
You will need to download the new a6p4data.scm file, and
will probably want the new a6code.com support code file.
Changes are described on the A6 page.
- [11/16] Quiz 6-9 statistics and normalization constants are now
up on the "Grading information" page (link above).
- [11/16] The snorkel data narrative is up on the A6 web page.
- [11/15] Decision tree slides have been posted to WebCT
- [11/15] Quiz 10 topics: Expect 1 or 2 questions from Quiz 9. The
main topic for Quiz 9 is decision trees. I'd highly suggest doing
problem 1 from the assignment. I will post my decision tree slides
and a "narrative" for the snorkel example on the assignment 6 page
this afternoon. Decision trees are covered in the text in chapter 18
sections 18.3. You should also be familiar with the "background"
material on learning covered in 18.1-2, e.g. supervised
vs. unsupervised learning and the idea of a hypothesis space.
Decision tree topics include what we covered in class this past week:
basic algorithm, information gain as a heuristic for choosing the best
attribute, ideas of noise and overfitting and 3 strategies for dealing
with noise and overfitting. Since chi-squared pruning is on the
assignment, this will not be on the quiz, but there may be questions
on the other two techniques.
- [11/10] Our class representatives are: Stacy (schoes2),
Jonathan (gryakj), Joe (schlij), and Rob (vandyr). At their request, I
have not given "email addresses" but only RCS IDs. You can figure it out...
- [11/8] Quiz 9 topics: Expect 1 question from Quiz 8. The main
topics for this quiz are: the relationship between gradient descent and
artificial neural network training (perceptron learning or
backpropagation), three layer feed-forward networks of sigmoid units,
stochastic gradient descent (not to be confused with the "stochastic
gradient method" on p. 742 of the text), overall context of perceptron
learning and backpropagation formulas. You should be able to apply
the backpropagation algorithm to an example. This material is mostly
covered in section 20.5 of the text, though I will put my slides on
WebCT and the handout from Thursday on the web page (handouts
section). (These should be up early this afternoon as I have still a
little work to do on them.)
Here are the formulas that will appear on the quiz (exactly as they
will appear)
- [10/30] Quiz 8 topics: perceptrons and neural networks --- what
they are, how they work. Know what kinds of functions a perceptron
can learn and be able to apply the perceptron learning algorithm.
Know the basics of neural networks: kinds of units, topology. How is
an artificial neural network different from animal neurons/brains?
This quiz will not cover the backpropagation algorithm.
- [10/25] Quiz 7 topics: a resolution proof using the set of
support strategy, definitions ("intuitive" and precise) of soundness
and completeness. Know what (Goedel's) completeness and
incompleteness theorems are about. Translating sentences from English
to FOL and from (general) FOL into CNF. Know completeness properties
for modus ponens (complete for Horn knowledge bases) and for
resolution (refutation complete).
- [10/17] I thought of a way for you to test your alpha-beta
minimax procedures on the written alpha-beta minimax from problem 3.
see the assignment 4 information page for details.
- [10/17] Quiz 6 topics: Expect 1-3 questions from Quiz 5 topics.
The main topics for Quiz 6 are the resolution inference rule,
conjunctive normal form (CNF), resolutions proofs in propositional
logic, first order logic (FOL), tranforming FOL sentences to CNF,
unification, forward and backward chaining in FOL. Be familiar with
sections 7.6 (resolution in prop. logic); 8.1 through 8.2 (basics of
FOL); and 9.1 through 9.4 (inference in FOL --- though only as
much as we've covered in class, i.e. basics of inference in FOL logic,
as we have not yet covered the material to the depth in chapter 9)
- [10/15] The A4p4 web tester is now up. The link is on the
Assignment 4 information page. Sorry for the delays.
The deadline for this problem is extended to 10:00am Monday October 20
with a first period late deadline of Tuesday 5pm and a second tier
late deadline of Wednesday 5pm.
- [10/15] There have been problems with the cgi-ai web
server, so the A4p4 web tester is not up yet. Hopefully everything
will be resolved today. I will extend the deadline for A4p4 by at
least a day, perhaps more depending on when it's all up and running.
- [10/11] The Connect 4 pre-tournament web stuff is up. See the
link on the above left. Hopefully we fixed all the bugs!
- [10/11] Quiz 5 topics: Expect 1-2 questions on game playing
search. The main topic for Q5 is logic, the material that we covered
last week in class. This includes: entailment, inference, soundness,
completeness, models, interpretations, truth tables, propositional
logic, proofs in propositional logic, forward chaining in
propositional logic, horn normal form. My slides from last week and
previous quiz solutions are available on WebCT. Be familiar with
Chapter 7: everything through section 7.4 and the beginning and
"forward [and backward] chaining" parts of section 7.5. As I said in
class, it is quite possible I will use the Wumpus world example on the
quiz.
- [10/8] There is now a page with grading information. Right now
it just has statistics on the quizzes.
- [10/7] The A4p5 web tester is up. Also, there are some annotated
games and a compiled version of my ab-minimax solutions available for
you to use in testing your evaluation function. See the Assignment 4
information page for details.
- [10/4] Quiz 4 topics: Expect 1 question on heuristic search (in
particular, on heuristics). Expect 1 or 2 questions on constraint
satisfaction problems. However, the main topic for Q4 is game playing
search: perfect and imperfect decisions, the MINIMAX search,
evaluation functions, alpha-beta pruning, EXPECTIMAX, extension to
multiplayer games. You should be familiar with Chapter 6.
- [9/29] The A4p12 web tester is now up.
- [9/29] There is a new version of a4code.com (version
1.2.1) which includes an implementation of corresponding-min
from Assignment 2.
- [9/25] Quiz 3 topics: Constraint satisfaction problems. Be
familiar with Chapter 5. Also, starting with this quiz there will be
some questions from the material from previous quizzes. Expect 1
question from Q1 topics and maybe 1 or 2 questions from Q2 topics
(particularly from those topics that weren't covered on Q2). Also add
"local beam search" to the list of iterative improvement search topics
from Q2.
- [9/25] The A3p8 web tester is up. Please note that this web
tester does not do any online tests but simply performs the syntax
check and pretest.
Since this web tester was up late, I will extend the deadline for
code submission for problem 8 (but not for the written part). Please
upload your code by midnight Thursday night. If for some reason this
is a hardship, email aistaff@cs.rpi.edu.
- [9/23] The final exam is scheduled for Tuesday December 16, 6:30
- 9:30 pm.
- [9/9] Quiz 2 topics: Heuristic search, heuristics, A* search,
Greedy search, admissibility, consistency (i.e. monotonicity),
automatically generating heuristics with relaxed problems, iterative
improvement algorithms: hill climbing, random restart hill climbing,
simulated annealing. Be familiar with Chapter 4 through Section 4.2
plus parts of Section 4.3.
- [9/9] Quiz 1 topics: Structuring problems for search, 6 blind
searches, completeness, optimality, time and space complexity,
avoiding repeated states. Be familiar with Chapter 3 through Section
3.5