Announcements
Homework 1 is available on the course website
Compass: compass2g.illinois.edu/ + navigate
Website: courses.engr.illinois.edu/cs440/
Piazza:
piazza.com/illinois/spring2017/cs440ece448/home
TA office hours starting next week in 207SC (basemen
Homework 3 Solutions
1. There are several combinations of the parameters that are acceptable. = 0.4, = 0.95, =
0.04 worked pretty well but since it depends a lot on implementation, we will accept numbers if
the justification looks right. For the number of
CS440 HW5 Solutions
3/16/2017 10:33:52 AM
Problem 1:
1) Is this a polytree? If so, how do you know?
If not, why not?
Yes, it is a polytree, because there is at most
one undirected path between two nodes.
2) How many parameters are necessary to
fully expre
Part 2: Text Document Classification
This part is to implement Naive Bayes classifiers to solve text classification problem.
The two main algorithms used here is Multinomial Naive Bayes and Bernoulli Naive
Bayes. Given the training data, how to achieve hi
Probability
Rational better theorem:
An agent who holds beliefs inconsistent with axioms of probability can be convinced
to accept a combination of bets that is guaranteed to lose them money.
o Random variables
Outcome: is a particular setting of all the
Part 1: Q-Learning (Pong) (for everybody)
Setup for Single-Player Pong (for everybody)
Before implementing the Q-learning algorithm, we must first write the code
for the gameplay of a single-player version of Pong. If you plan on doing
the 4-unit assignme
Machine learning
1. Generalization and overfitting
Want classifier does well on never seen before data
Good performance on the training/validation set, poor performance on the test set
2. NN VS linear
-Pros: 1. Simple to implement 2. Decision boundaries n
Bayes Nets: (1) Review (2) Semantics (3) Burglars (4) Concepts (5) Design (6) Research
Review: Bayesian inference
A general scenario:
Query variables: X
Evidence (observed) variables and their values: E = e
Inference problem: answer questions about the qu
CS440/ECE448 Fall 2016 Final Review Solutions
1. Use the axioms of probability to prove that P(A) = 1 P(A).
Solution
Summary of axioms:
i)
P(A) 0 for all events A;
ii)
P()=1;
iii)
If AB = empty then P(A U B) = P(A) + P(B).
P(A)+P(A) = P(A U A) = P() = 1 (
Bayesian Inference &
Nave Bayes
10/17/2016
Review: Probability
Random variables, events
Axioms of probability
Atomic events
Joint and marginal probability distributions
Conditional probability distributions
Product rule, chain rule
Independence and condit
Bayes network inference
A general scenario:
Query variables: X
Evidence (observed) variables and their values: E = e
Unobserved variables: Y
Inference problem: answer questions about the query
variables given the evidence variables
This can be done using
Markov Decision Processes
(Chapter 17)
Image source: P. Abbeel and D. Klein
Markov Decision Processes
In HMMs, we see a sequence of observations and
try to reason about the underlying state sequence
There are no actions involved
But what if we have to
Name and Netid _
CS440/ECE448 Fall 2016 Midterm (Out of 50 points, max 52 possible)
1. Mark each statement as True or False, and give a brief explanation. (16 points)
2 points each (1 for T/F, 1 for explanation)
a. An environment can be fully observable a
Assignment 4 grading instructions
Total for 3 units: 20 points
Total for 4 units: 10 points
Code or report missing: give 5 points for the entire assignment.
If part of the assignment was submitted before the deadline with additional
parts after the d
Homework 4
Not collected; Not graded; For practice only
1. Frequentist and Bayesian statisticians sometimes agree but sometimes disagree on what constitutes a well formed
statistical situation. For each statement below explain why a frequentist and a Baye
Homework 2 Solutions
Assigned 1/31/17
Due 2/7/16
1. Suppose we could destroy any block in the otherwise traditional blocks world. The destruction is gentle with any
supported blocks settling down on the support under the destroyed block.
a) Define the ope
HW1 due Tuesday 1/31
Our First (trial) Q/A Sessions
(completely optional)
Monday 1/30/17; Room 4405 SC
9-10AM
12-1PM
6-7PM
7-8PM
We may lose the room (another facet of registration problem)
If we lose the room, look for a note on the door for a new
Announcements
HW3 (with MP) is available
Two extra credit opportunities
Earlier one
(up to 15% bonus)
Standard one
(up to 20% bonus)
You CAN collect both
But do NOT hand in more than one solution to any
part of HW3
There is support code available w
Announcements
The Monday Q/A for HW3 will be 2/20 (not this
Monday)
In general: the Monday just prior to the Tuesday due
date
HW1 Issue: Convex Sets
A set S is convex iff
x,y S, t [0,1] (1-t)x + ty S
The straight line between any two points in the se
Announcements
Midterm exam March 1, one week from
tomorrow
Topics:
Models
Search & Optimization
Logic, Planning, Knowledge Representation
Reinforcement Learning
Statistics
Practice homework HW4 is out
Solutions for HW3 will be posted soon
The Two Enve
Announcements
Reminder: Midterm Exam
March 1 (two weeks from yesterday)
6:30PM
Exam rooms will be assigned
You must take the exam in your assigned room for it
to count
HW4 will be on statistics
Available next week
It will not be collected or graded
M
Estimation
An estimator for a quantity is a function from
samples to a value
It makes a guess based on the data
We might estimate the mean of a distribution
from the average of some samples
Simple but can be problematic
take a coin from your pocket,
Can we ever stop
learning / exploring?
We try a2 in state s7 after many tries, it doesnt look so
good
Can we ever stop trying action a2 in state s7
Question formalized as the two-arm bandit problem
(costs nothing to play!)
Arm1 pays $1 with probability p
Announcements
Midterm Exam
6:30 7:45 PM Wed. March 1
A-M If the first letter of your NetID is A-M
take the exam here 1320DCL
N-Z
If the first letter of your NetId is N-Z
take the exam Loomis 151 (Physics building)
We only have just enough seats so yo
CS 440 / ECE 448 Spring 2017
Introduction to Artificial Intelligence
Prof. Gerald DeJong, [email protected]
3320SC
TAs:
Daniel Calzada
Codruta Girlea
Akshat Gupta
Kyo Kim
See website (soon) for Syllabus, TA emails and office hours
1
Official announcemen
Homework 4
Not collected; Not graded; For practice only
1. Frequentist and Bayesian statisticians sometimes agree but sometimes disagree on what constitutes a well formed
statistical situation. For each statement below explain why a frequentist and a Baye
CS 440 HW 6
Extra Credit Due Date: April 2, 2017 11:59 pm
Due Date: April 4, 2017 11:59 pm
1. We have a dataset of 30 students in which each student has 3 features, namely BMI, Grade,
and PlayBall. BMI has 2 possible values: normal (N) and overweigh
Homework 1
Assigned 1/19/17
Due 1/31/17
Consider the search tree below. For each search strategy below, a) give the goal state found (if none, write
NONE), b) say whether or not this goal state is optimal, and c) give the order in which the nodes are visi