Topic Outline for Final Exam
CS 3600 Intro to Artificial Intelligence
Prof. Jim Rehg
School of Interactive Computing
Georgia Institute of Technology
Topic Outline1
1. Agents (Chpts 1 & 2)
Agents as functions
Rational agent behavior
Task environments an
History
1200s: Automatons
1770: Mechanical turk
1800s: The term computer
Babbage Analytic Engine
1950: Turing Test
1956: Term Articial intelligence coined at
Dartmouth Conference
Cognitive revolution: Chomsky, 1955
Previously: behaviorism
sensory
behavior
What an agent needs
Sensors
Effectors
Performance measure
Success criterion
What you are trying to achieve
e.g., maximize the performance measure
House stays empty
People enter house randomly
People enter randomly, costs 1 point to move
Dont have a map
Se
Search!
Goal-based problem-solving agents
State: unique conguration of the relevant facts
about the world
Goal: State you want to be in
Problem: You dont know how to get from your
current state to a goal state
When to use search: when you dont know how to
Evaluating search strategies
Completeness
Does it always nd a solution
Can it visit all states
Time complexity
Space complexity
Number of states generated
Number of states stored in memory at any given time
Optimality
Does it always nd the least cost?
ops
Uniform Cost Search
(pretend nothing west of S)
Uniform Cost Search
What if actions have costs?
Instead of shortest (# actions), we want cheapest
sequence
g(n) : the cost of traversing from initial state to state
n via the shortest known path
WHILE (NOT i
A*
Admissibility
Admissible heuristic: Guaranteed never to
overestimate the cost of reaching the nearest goal
Let f*(n) = g*(n) + h*(h) be the perfect estimate
a heuristic h(n) is admissible if:
for all n, h(n) h*(n)
Informedness
informedness:
size of to
Minimax
Function MINIMAX (state)
(v, a) = max-value(state)
return a
Function MAX-VALUE (state)
IF terminal(state) THEN return utility(state)
v = -
FOREACH (a, s) in successors DO
v = max(v, min-value(s) )
a = action that corresponds to v
return (v, a)
Fun
import copy
import sys
from datetime import datetime
from math import exp
from random import random, randint, choice
class Perceptron(object):
"
Class to represent a single Perceptron in the net.
"
def _init_(self, inSize=1, weights=None):
self.inSize = i
Project 1: Search in Pacman
All those colored walls,
Mazes give Pacman the blues,
So teach him to search.
Introduction
In this project, your Pacman agent will find paths through his maze world, both to reach a particular
location and to collect food effic
Apr 29,2015
CS 3600 Final Exam
Name/GTID:
Question 1. Naive Duck
your friend Howard th.e Duck has constructed a highly-biased probabilistic model of
the animal kingdom, consisting of the joint probability distribution P(W,Q, D), where
I4l standS for "walk
Name/GTID:
CS 3600 Practice Final Exam
Apr 22, 2016
Table 1: For instructors use
Question
Points Scored
Possible Points
1
12
2
12
3
6
4
12
5
12
6
6
7
12
8
2-4
The final exam is closed book. You are allowed 4 sheets of notes (8 pages front and back).
You m
Project 4: Decision Trees
Introduction
This project is intended to familiarize you with one of the standard approaches to classification
problems, decision trees. You will code up decision tree learning and then apply it to several
relatively simple probl
CS 3600 Midterm Practice Exam
Feb 19,2016
Question L. Use A* Search to solve the map problem below, where the numbers give the
distances between states.
r
Initial State = A
o Goal State = E
o The value of the heuristic function for each node
n
A
h(")
B
C
Project 5: Neural Nets
Introduction
In this project you will be implementing neural nets, and in particular the most common
algorithm for learning the correct weights for a neural net from examples. Code structure is
provided for a Perceptron and a multi-
Project 3: Ghostbusters
I can hear you, ghost.
Running won't save you from my
Particle filter!
Introduction
Pacman spends his life running from ghosts, but things were not always so. Legend has it that many years
ago, Pacman's great grandfather Grandpac l
Consistency
Shortcuts happen when h overestimates a states
distance to nearest goal
A heuristic may also grossly underestimate
Consistent heuristic: h(A) - h(B) K(A,B) for all A,B
K(A,B) is the actual cost of transitioning from A to B
(dont need to be nei
Start
Rook Jumping Maze
Solu%on: DRLUDLRULLRDU
Start
Generate maze
state?
init?
goal?
Instructions: From each numbered square, one may move that exact number of
squares horizontally or vertically in a straight line. Starting at the circled square
in th
Game Playing
Games as search
Fully-observable
Deterministic (until we add dice)
Static
Multi-agent adversarial
Zero-sum games
Turn-taking
1997: Deep Blue beat
world champion.
Deep Blue searches 200
million positions/second.
Look-ahead up to 40-ply
2007: S
Uncertainty and Probability, Lecture 1
Lecture 15
Jim Rehg
February 13, 2017
Based in part on lecture
slides from Mark Paskin
College of Computing
Georgia Tech
Uncertainty
We need to be able to describe propositions which are not exactly
true or false
pne
Constraint Satisfaction, Part 1
Lecture 7
Chapter 6, Sections 6.1-6.3
Jim Rehg
College of Computing
Georgia Tech
January 25, 2016
Search So Far
Classical (combinatorial) search:
State space has no explicitly represented structure
Implicit structure in gen
Constraint Satisfaction, Part 2
Lecture 8
Chapter 6, Sections 6.1-6.3
Jim Rehg
College of Computing
Georgia Tech
January 27, 2016
Ways to Improve Backtracking Search
What variable to assign next?
Choose variable with fewest values left fail first
What ord
A* Search
Lecture 5
Informed Search Algorithms (Ch 3.5)
Problem Solving Agents (Search) Lecture 3
Jim Rehg
A* Search
Most widely known Best-First
alg
f(n) = g(n) + h(n)
g(n) = cost to get to this
node
h(n) = estimated cost from
here
Minimizes total soluti
Uncertainty and Probability, Lecture 2
Lecture 16
Jim Rehg
February 17, 2017
Based in part on lecture
slides from Mark Paskin
College of Computing
Georgia Tech
Joint Densities for Random Variables and
Discrete case joint pdf is a table:
false
true
false
0
Search, Part 1
Lecture 3
Chapter 3, Sections 3.2-3.3
Jim Rehg
College of Computing
Georgia Tech
January 13, 2017
Problem Types
Assumptions for now:
- Observable state
- Static environment
- Discrete states/actions
- Deterministic actions
Intro to AI, Geor
Agents and Environments
Lecture 2
Jim Rehg
January 11, 2017
College of Computing
Georgia Tech
(Some figures due to Andrea Thomaz)
Admin Update
The lecture schedule in T-Square syllabus is now complete
This schedule is subject to change
Any changes announc
Local Search and Optimization
22c:31:002 Algorithms
Outline
Local search techniques and optimization
Hill-climbing
Gradient methods
Simulated annealing
Genetic algorithms
Issues with local search
Local search and optimization
Previously: systematic explo
CS 2110
Timed Lab 1
Due Date and Time
Day:
Wednesday, February 8th
Time: Before the end of your assigned lab section
Policy
Submission
TURN IN THIS ASSIGNMENT ELECTRONICALLY USING T-Square. SUBMISSIONS
WHICH ARE LATE WILL NOT BE ACCEPTED. EMAIL SUBMISSION