Article Directory
Agent and rational
Agent: sensing by the sensor is generated by the action of the environment and the environment of the actuator Agent
Agent sensing sequence: the Agent received a complete history of all input data is the
perceptual information: Sensory Input Agent
sensing sequence: sensing information collection
Agent functions: description Agent behavior, the perception of any given action sequence mapped to
Agent program: Implementing Agent function
The agent function describes what the agent does in all circumstances
The agent program calculates the agent function
The concept of rationality
Reason: an attribute Agent, taking into account the perception so far, they choose to maximize their expected utility behavior.
Reason (Rationality) Agent: For each possible sequence of perception, based on prior knowledge of the sequence of known evidence provided and intelligent body construction, a rational agent should be able to select the desired action to maximize performance.
Rationality depends on:
- The performance measure that defines the criterion of success
- The agent’s prior knowledge of the environment
- The actions that the agent can perform
- The agent's percept sequence to date
performance metrics, prior knowledge, executable action, perception sequence
Knowing Agent: Agent omniscient clearly know the actual results of his actions generated and make the appropriate action, in reality impossible. Rational ≠ omniscience
perfect Agent: Agent make rational expectation maximize performance; Perfect Agent actual performance is maximized. ≠ perfectly rational
Rational Agent should also be autonomous , it should learn to compensate for incomplete or incorrect prior knowledge.
PEAS and environmental characteristics
Specification task environment comprises:
the To A Rational Design WE need to Agent A the Specify Task Environment
- Performance measure
- Environment
- Actuators
- Sensors
PEAS (Performance Performance, environmental Environment, actuators Actuators, Sensors Sensors)
For example, Spam Filtering spam filtering
- Performance measure: spam block
- Environment: email client or server
- Actuators: mark as spam, transfer messages
- Sensors: emails (possibly across users), etc.
Task environmental attributes
- Fully observable and partially observed
Sensors Agent is able to get the full state of the environment at each time point. If the monitored information related to the decision-making action, the task environment is fully observable effective. - Single Agent and Multi-Agent
- Determined and random
- Fragmented and continuity style
segments: The next segment is not dependent on the actions taken by the previous segment (such as most of the classification task).
Continuous: the current decision will affect all future decisions. - Static and dynamic
depending on whether the environment will change when Agent calculations. - Discrete and continuous
state of the environment, treatment time, Agent sensory information and action are discrete and continuous points.
Agent structure
Agent = architecture + program (Agent program)
architecture, a physical computing device having sensors and actuators
agent program, each received a new perceptual information, which will be added to the perception of the sequence, and in accordance with prior knowledge to give a correspondence table operation to map the sensing information from the sensor to the action function Agent
All agents have the same skeleton
- Input = current percepts
- Output = action
- Program= manipulates input to produce
output
Four basic types in order of increasing generality
- Simple reflex agents simply reflected Agent , ignore the perceived sequence, but only for the currently selected action perception.
- Model-based reflex agents based agent model
- Goal-based agents based on agent targets
based on the target's agent will consider the consequences of actions and the actions to be taken, namely how far the target - Utility-based agents based utility Agent , maximize expected
- Learning agents 学习agent
All these can be turned into learning agents
How components Agent program running
The work of the member
- Atom representation
- Feature representation (state in which feature vectors)
- Structured representation (+ feature vector relationship to other objects)
summary
Recalling the main points such as Bu:
Agent is something you can sense the environment and act in the environment.
Agent Agent designated function in response to any action taken by the perceived sequence.
Agent performance metrics evaluation of performance in the environment behavior . Agent sequence given to the perception of rational action to pursue Agent performance metrics to maximize the expected value .
Task environment includes specification of a performance metric, the external environment, actuators and sensors . When designing Agent, the first step is always the space mission was defined as complete as possible.
Task environment there are many variations to see from a different dimension. They may be completely or partially observed, mono- or Agent Agent, deterministic or random, fragmented or a continuation formula, static or dynamic, continuous or discrete, known and unknown .
Agent Agent program is to achieve the function. There are various basic Agent program design reflects the explicit table
now and the kind of information for decision-making process. Design may be efficient, changes in compressibility and flexibility. Agent proper design program depends on the nature of the environment.
Simple reflection Agent directly respond to sensory information. Agent states to maintain internal reflection-based model, its operation model directly from within the current state of the world is derived and updated over time. Agent selection based on the target of action to achieve its objectives indicated by the display, and utility-based Agent tries to maximize its expected utility can select the action.
All Agent can by learning to improve their performance.
Information sharing
Experiment download:
https://github.com/yyl424525/AI_Homework
Artificial Intelligence - A Modern Approach Chinese third edition pdf, courseware, operations and solutions, after-school exercise answers, test code and reports, over the years Kaobo title Download: HTTPS : //download.csdn.net/download/yyl424525/11310392