Belief-optimal Reasoning for Cyber
Download
Report
Transcript Belief-optimal Reasoning for Cyber
CS B551: Elements of Artificial
Intelligence
Instructor: Kris Hauser
http://cs.indiana.edu/~hauserk
1
Recap
http://cs.indiana.edu/classes/b551
Brief history and philosophy of AI
What is intelligence? Can a machine
act/think intelligently?
• Turing machine, Chinese room
2
Agenda
Agent Frameworks
Problem Solving and the Heuristic
Search Hypothesis
3
Agent Frameworks
4
Definition of Agent
Anything that:
• Perceives its environment
• Acts upon its environment
A.k.a. controller,
robot
5
Definition of “Environment”
The real world, or a virtual world
Rules of math/formal logic
Rules of a game
…
Specific to the problem domain
6
Agent
Percepts
Sensors
Actions
Actuators
Environment
?
7
Agent
Percepts
Sensors
Actions
Actuators
Environment
?
Sense – Plan – Act
8
“Good” Behavior
Performance measure (aka reward,
merit, cost, loss, error)
Part of the problem domain
9
Exercise
Formulate the problem domains for:
• Tic-tac-toe
• A web server
• An insect
• A student in B551
• A doctor diagnosing
a patient
• IU’s basketball team
• The U.S.A.
What is/are the:
• Environment
• Percepts
• Actions
• Performance measure
How might a “goodbehaving” agent process
information?
10
Types of agents
Simple reflex (aka reactive, rulebased)
Model-based
Goal-based
Utility-based (aka decision-theoretic,
game-theoretic)
Learning (aka adaptive)
11
Simple Reflex
Percept
Interpreter
State
Rules
Action
12
Simpl(est) Reflex
State
Observable
Environment
Interpreter
State
Rules
Action
13
Simpl(est) Reflex
State
Observable
Environment
Rules
Action
14
Rule-based Reflex Agent
A
B
if DIRTY = TRUE then SUCK
else if LOCATION = A then RIGHT
else if LOCATION = B then LEFT
15
Model-Based Reflex
Percept
Interpreter
State
Rules
Action
Action
16
Model-Based Reflex
Percept
Model
State
Rules
Action
Action
17
Model-Based Reflex
Percept
Model
State
State
estimation
Rules
Action
Action
18
Model-Based Agent
State:
LOCATION
HOW-DIRTY(A)
Rules:
if LOCATION = A then
if HAS-SEEN(B) = FALSE then RIGHT
HOW-DIRTY(B)
else if HOW-DIRTY(A) > HOW-DIRTY(B) then SUCK
HAS-SEEN(A)
HAS-SEEN(B)
A
else RIGHT
…
B
Model:
HOW-DIRTY(LOCATION) = X
HAS-SEEN(LOCATION) = TRUE
19
Model-Based Reflex Agents
Controllers in cars, airplanes,
factories
Robot obstacle avoidance, visual
servoing
20
Goal-Based, Utility-Based
Percept
Model
State
Rules
Action
Action
21
Goal- or Utility-Based
Percept
Model
State
Decision Mechanism
Action
Action
22
Goal- or Utility-Based
State
Decision Mechanism
Action Generator
Percept Model
Model
Simulated State
Performance tester
Best Action
Action
23
Goal-Based Agent
24
Big Open Questions:
Goal-Based Agent = Reflex Agent?
Physical Environment
“Mental Environment”
Percept
Model
Mental Model
State
Mental State
DM Rules
Action
Action
Mental Action
25
Big Open Questions:
Goal-Based Agent = Reflex Agent?
Physical Environment
“Mental Environment”
Percept
Model
Mental Model
State
Mental State
DM Rules
Action
Action
Mental Action
26
With Learning
Percept
Model/Learning
State/Model/DM specs
Decision Mechanism
Action
Action
27
Big Open Questions: Learning Agents
The modeling, learning, and decision
mechanisms of artificial agents are
tailored for specific tasks
Are there general mechanisms for
learning?
If not, what are the limitations of the
human brain?
28
Types of Environments
Observable / non-observable
Deterministic / nondeterministic
Episodic / non-episodic
Single-agent / Multi-agent
29
Observable Environments
Percept
Model
State
Decision Mechanism
Action
Action
30
Observable Environments
State
Model
State
Decision Mechanism
Action
Action
31
Observable Environments
State
Decision Mechanism
Action
Action
32
Nondeterministic Environments
Percept
Model
State
Decision Mechanism
Action
Action
33
Nondeterministic Environments
Percept
Model
Set of States
Decision Mechanism
Action
Action
34
Agents in the bigger picture
Binds disparate fields
(Econ, Cog Sci, OR,
Control theory)
Framework for technical
components of AI
• Components are useful and
rich topics themselves
• Rest of class primarily
studies components
Casting problems in the
framework sometimes
brings insights
Agent
Robotics
Reasoning
Search
Perception
Learning
Knowledge Constraint
rep.
satisfaction
Planning
Natural
language
...
Expert
Systems
35
Problem Solving and the
Heuristic Search Hypothesis
36
Example: 8-Puzzle
8
2
3
4
5
1
1
2
3
7
4
5
6
6
7
8
Initial state
Goal state
State: Any arrangement of 8 numbered tiles and an empty tile on a 3x3 board
37
Successor Function: 8-Puzzle
2
3
4
5
1
6
8
2
7
8
6
3
8
2
3
4
7
3
4
5
1
6
5
1
7
SUCC(state) subset of states
8
The successor function is knowledge
about the 8-puzzle game, but it does
not tell us which outcome to use, nor to
which state of the board to apply it.
5
2
7
4
1
6
38
Across history, puzzles and games
requiring the exploration of alternatives
have been considered a challenge for
human intelligence:
Chess originated in Persia and India
about 4000 years ago
Checkers appear in 3600-year-old
Egyptian paintings
Go originated in China over 3000 years
ago
So, it’s not surprising that AI uses games
to design and test algorithms
39
Exploring Alternatives
Problems that seem to require
intelligence require exploring
multiple alternatives
Search: a systematic way of
exploring alternatives
40
Defining a Search Problem
S
State space S
Successor function:
x S SUCC(x) 2S
Initial state s0
Goal test:
xS GOAL?(x) =T or F
Arc cost
41
Problem Solving Agent
Algorithm
1. I sense/read initial state
2. GOAL? select/read goal test
3. SUCC select/read successor
function
4. solution search(I, GOAL?, SUCC)
5. perform(solution)
42
State Graph
Each state is
represented by a
distinct node
An arc (or edge)
connects a node s
to a node s’ if
s’ SUCC(s)
The state graph may
contain more than one
connected component
43
Solution to the Search Problem
A solution is a path
connecting the initial
node to a goal node
(any one)
The cost of a path is
the sum of the arc
costs along this path
An optimal solution is
a solution path of
minimum cost
There might be
no solution !
G
I
44
Pathless Problems
Sometimes the path
doesn’t matter
A solution is any goal
node
Arcs represent
potential state
transformations
E.g. 8-queens,
Simplex for LPs, Map
coloring
G
I
45
8-Queens Problem
State repr. 1
• Any nonconflicting
placement of
0-8 queens
State repr. 2
• Any placement
of 8 queens
46
Intractability
It may not be
feasible to
construct the
state graph
completely
n-puzzle:
(n+1)! states
k-queens:
kk states
47
Heuristic Search Hypothesis
(Newell and Simon, 1976)
“The solutions to problems are represented as symbol
structures. A physical symbol system exercises its intelligence
in problem solving by search - that is, by generating and
progressively modifying symbol structures until it produces a
solution structure.”
Intelligent systems must use heuristic
search to find solutions efficiently
Heuristic: knowledge that is not presented
immediately by the problem specification
48
Example
I’ve thought of a number between 1
and 100. Guess it.
49
Example
I’ve picked a password between 3
and 8 alphanumeric characters that
I’ll never forget. Guess it.
50
Discussion
Debated whether all intelligence is
modifying symbol structures…
• e.g., Elephants don’t play chess, Brooks ’91
But for those tasks that do require
modifying symbol structures,
hypothesis seems true
• Perhaps circular logic?
51
Topics of Next 3-4 Classes
Blind Search
• Little or no knowledge about how to
search
Heuristic Search
• How to use heuristic knowledge
Local Search
• With knowledge about goal distribution
52
Recap
Agent: a sense-plan-act framework
for studying intelligent behavior
“Intelligence” lies in sophisticated
components
53
Recap
General problem solving framework
• State space
• Successor function
• Goal test
• => State graph
Search is a methodical way of
exploring alternatives
54
Homework
Register!
Readings: R&N Ch. 3.1-3.3
55
What is a State?
A compact representation of
elements of the world relevant to
the problem at hand
• Sometimes very clear (logic, games)
• Sometimes not (brains, robotics, econ)
History is a general-purpose state
representation: [p1,a1,p2,a2,…]
State should capture how history
affects the future
56