Вы находитесь на странице: 1из 20

Advanced

Intelligent
Systems

Thomas
Anung Basuki

Concepts Advanced Intelligent Systems


Rationality
Second Slide, Intelligent Agents
Environment

Agents

Thomas Anung Basuki

August 26, 2016


Table of Contents

Advanced
Intelligent
Systems

Thomas
Anung Basuki
1 Agents and Environments
Concepts

Rationality

Environment 2 Rationality
Agents

3 Environment

4 Classes of Agents
Agents and Environments

Advanced
Intelligent
Systems Agent: perceiving environment through sensors and acting through actuators
Thomas Percept: agent’s perceptual inputs at any given instant
Anung Basuki Percept Sequence: a complete history of percepts
Agent function: maps percept sequence to an action
Concepts

Rationality

Environment

Agents
Vacuum Cleaner World

Advanced
Intelligent
Systems Two locations: A and B
One out of two conditions on each location: Dirty or Clean
Thomas
Anung Basuki Three possible actions: Suck, Right, Left
Agent behaviour is defined by filling a table of agent function
Concepts

Rationality

Environment

Agents
Table of Contents

Advanced
Intelligent
Systems

Thomas
Anung Basuki
1 Agents and Environments
Concepts

Rationality

Environment 2 Rationality
Agents

3 Environment

4 Classes of Agents
Rationality

Advanced
Intelligent
Systems

Thomas
Anung Basuki
is defined in terms of agent performance in achieving a
Concepts
goal
Rationality
The performance is obtained by observing environment
Environment

Agents
states
Rationality depends on
performance measure that defines the criterion of success
the agent’s prior knowledge of environment
possible actions the agent can perform
percept sequence
Rational Agent

Advanced
Intelligent
Systems

Thomas
Anung Basuki
For each possible percept sequence
Concepts
should select an action
Rationality
expected to maximise its performance measure
Environment
based on percept sequence and built-in knowledge
Agents
is not omniscience (know the actual result of actions)
sometimes require information gathering or even exploration
learning to improve knowledge
Autonomy is measured from how much the agent depends on percepts than
prior knowledge
Back to Vacuum Cleaner

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts awards 1 point for each clean square at each step (the
Rationality lifetime is 1000 steps)
Environment the geography of environment is known except the dirt
Agents
distribution
clean squares stay clean and sucking cleans a dirt square
Left and Right moves the agent only if it is still in the
environment
Table of Contents

Advanced
Intelligent
Systems

Thomas
Anung Basuki
1 Agents and Environments
Concepts

Rationality

Environment 2 Rationality
Agents

3 Environment

4 Classes of Agents
The Nature of Environment

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts PEAS: Performance Environment Actuators Sensors


Rationality Task environment:
Environment fully observable, partially observable, unobservable
Agents single agent or multiagent (competitive vs. cooperative)
deterministic vs. stochastic
episodic vs. sequential
static vs. dynamic
discrete vs. continuous
known vs. unknown
Environment Examples

Advanced
Intelligent
Systems

Thomas Environment Obs. Agents Det. Episodic Static Discrete


Anung Basuki
Crossword Puzzle F Single Det. Seq. Static Discrete
Concepts
Chess with clock F Multi Det. Seq. Semi Discrete
Rationality
Poker P Multi Stoc. Seq. Static Discrete
Environment
Backgammon F Multi Stoc. Seq. Static Discrete
Agents
Taxi driving P Multi Stoc. Seq. Dynamic Continuous

Medical diagnosis P Single Stoc. Seq. Dynamic Continuous

Image analysis F Single Det. Epi. Semi Continuous

Part-picking robot P Single Stoc. Epi. Dynamic Continuous

Refinery Controller P Single Stoc. Seq. Dynamic Continuous

English Tutor P Multi Stoc. Seq. Dynamic Discrete


Table of Contents

Advanced
Intelligent
Systems

Thomas
Anung Basuki
1 Agents and Environments
Concepts

Rationality

Environment 2 Rationality
Agents

3 Environment

4 Classes of Agents
The Structure of Agents

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts
agent = architecture + program
Rationality

Environment Table-driven agents


Agents Reflex agents
Goal-based agents
Learning agents
Table-Driven Agent

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts Table-Driven-Agent(percept) /*percepts, a sequence, initially empty*/


Rationality /*table, a fully specified table of actions, indexed by percept sequences*/
append percept to the end of percepts
Environment action ← LOOKUP(percepts,table)
return action
Agents
needs a very large memory

and a very long time to look up the table


Simple-Reflex Agent

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts Simple-Reflex-Agent(percept) /*rules, a set of condition-action rules*/


state ← INTERPRET-INPUT(percept)
Rationality rule ← RULE-MATCH(state,rules)
action ← rule.ACTION
Environment return action

Agents
state is observed

observed state is used to find appropriate rule

action is part of the rule


Simple-Reflex Agent

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts

Rationality

Environment

Agents
Goal-Based Agent

Advanced
Intelligent
Systems

Thomas
Anung Basuki

Concepts

Rationality
State is not enough to decide which action to choose
Environment Especially when there are many options of action
Agents A goal-based agent always tries to select actions that
move him closer to the goal
Search and planning are used to find the correct actions
Introduction to Intelligent Systems
Agents

Goal-Based Agent
Introduction to Intelligent Systems
Agents

Learning Agent

One way to build intelligent agents is to build learning


machines and then teach them.
Divided into 4 components
performance element: the previously defined utility agent
learning element: responsible for making improvements
critic: responsible for providing feedbacks on the agent’s
performance
problem generator: generates problems for exploratory
experiments
Introduction to Intelligent Systems
Agents

Learning Agent

Вам также может понравиться