Artificial Neural Networks
An artificial neural network is a model that emulates a biological neural
network.
The concepts are used to implement software simulations of the
massively parallel process that involve processing elements (also
called artificial neurons or neuroses) interconnected in network
architecture.
The artificial neuron receives inputs that are analogous to the
electrochemical impulses that the dendrites of biological neurons
receive Ron other neurons.
The output of the artificial signals can be neuron corresponds to signals
sent out from a biological neuron over its axon.
Learning
An ANN learns from its experiences.
The usual process of learning involves three tasks.
1. Compute outputs
2. Compare outputs with desired targets
3. Adjust the weights and repeat the process
Neural Network Hardware
Most current neural network applications involve software simulations
that run on conventional sequential process (called simulated artificial
neural networks, or SANN).
To increase the computational speed when regular computers are
used, one of four approaches is applicable.
1. Faster general purpose computers
2. General purpose parallel processors
3. Neural chips
4. Acceleration boards
Faster general purpose computers
Use a high end PC with a fast Pentium, P6 or PowerPc processor chip
or a RISC workstation like an IBM RS/6000.
General purpose parallel processor
There are a number of general purpose parallel processors, ranging
from the large grained IMB SP – 2 to small grained transcomputers
such as INMOS Transputer.
Neural computing is inherently parallel in nature.
There are a number of parallel neural network codes such as Nnets
from Cosmic / The University of Georgia.
Neural chips
Most of today’s special chips can execute computations very fast, but
they cannot be used to train the network.
The idea of a chip is to provide implementation of neural network data
structures on the following.
An Analog chip _ such as Intel 80170NX Electronically Trainable ANN,
or ETANN
A Digital chip _ such as Micro Devices MD – 1220
An optical chip.
Acceleration boards
These are powerful, dedicated parallel / array processors that can be
installed in regular computers, similar to a math coprocessor.
Such a processor can be thousand soft times faster than a Pentium
processor.
Some examples are….
Adaptive Solutions CNAPS/PC–128 – Which BrainMakers uses
The Mosaic QED Board The HNC Balboa Board
Acceleration bards are extremely useful because they reduce training
time, which is usually quite long.
Intelligent Agents and Creativity
An intelligent Agent (IA) is a computer program that helps a user with
routine computer tasks.
The first vignette is an example of a first generation individually
operated agent.
They can overcome the most critical limitation of the internet – the
information overflow – and make electronic commerce a viable
organization tool.
Names
Several names are used to describe intelligent agents, including
software agnates, wizards, knowbots, and softbots (intelligent software
robots).
The names sometimes reflect the nature of the agent.
Characteristics of Intelligent Agents
Autonomy Learning
Operates in background Reactivity
Single task Proactive ness
Communication Temporal continuity
Automates repetitive tasks Personality
Supports conditional processing Mobility
Autonomy
An agent is autonomous; that is, it is capable of action on its own.
An agent must be able to make some decisions on its own by being
goal oriented, collaborative, and flexible.
It must b able to alter its course or behavior when it meets obstacles to
find ways around the impediment.
Example _ an agent should be able to accept high level request and
decide on its own where and how to carry out the request.
Autonomy implies that an agent takes initiative and exercises control
over its own actions in the following ways.
Goal oriented Accepts high level requests indicating what a human wants and
is responsible for deciding how and where to satisfy certain
requests.
Collaborative Does not blindly bet commands, but can modify requests, ask
clarification questions, or even refuse to satisfy certain requests.
Flexible Actions are not scripted, able to dynamically choose which
actions to invoke and it what sequence, in response to the state
of its external environment.
Self Starting Unlike standard programs directly invoked by user, an agent can
sense changes of its environment and decide when to act.
Operates in backboard
An agent must be able to work out of sight, within the realm of
cyberspace or other computer systems, without the constant attention
of its user.
Some developers use the term remote execution or mobile agents to
describe this attribute.
Single task
In most cases an agent is designed to accomplish a single task.
A single task cold is searching the internet for articles.
Another might be filtering electronic mail.
The more futuristic agents appear capable of doing multiple things.
Communication
Many agents are designed to interact with other agents, humans, or
software programs.
Instead of making a single agent smarter, additional agents can be
created to handle undefeated tasks.
The ability to collaborate enables the system to accomplish more
complex tasks.
Automates repetitive tasks
An agent is designed to do narrowly defined tasks, which it can do over
and over without getting bored or sick or going on strike.
An artificial neural network is a model that emulates a biological neural
network.
The concepts are used to implement software simulations of the
massively parallel process that involve processing elements (also
called artificial neurons or neuroses) interconnected in network
architecture.
The artificial neuron receives inputs that are analogous to the
electrochemical impulses that the dendrites of biological neurons
receive Ron other neurons.
The output of the artificial signals can be neuron corresponds to signals
sent out from a biological neuron over its axon.
Learning
An ANN learns from its experiences.
The usual process of learning involves three tasks.
1. Compute outputs
2. Compare outputs with desired targets
3. Adjust the weights and repeat the process
Neural Network Hardware
Most current neural network applications involve software simulations
that run on conventional sequential process (called simulated artificial
neural networks, or SANN).
To increase the computational speed when regular computers are
used, one of four approaches is applicable.
1. Faster general purpose computers
2. General purpose parallel processors
3. Neural chips
4. Acceleration boards
Faster general purpose computers
Use a high end PC with a fast Pentium, P6 or PowerPc processor chip
or a RISC workstation like an IBM RS/6000.
General purpose parallel processor
There are a number of general purpose parallel processors, ranging
from the large grained IMB SP – 2 to small grained transcomputers
such as INMOS Transputer.
Neural computing is inherently parallel in nature.
There are a number of parallel neural network codes such as Nnets
from Cosmic / The University of Georgia.
Neural chips
Most of today’s special chips can execute computations very fast, but
they cannot be used to train the network.
The idea of a chip is to provide implementation of neural network data
structures on the following.
An Analog chip _ such as Intel 80170NX Electronically Trainable ANN,
or ETANN
A Digital chip _ such as Micro Devices MD – 1220
An optical chip.
Acceleration boards
These are powerful, dedicated parallel / array processors that can be
installed in regular computers, similar to a math coprocessor.
Such a processor can be thousand soft times faster than a Pentium
processor.
Some examples are….
Adaptive Solutions CNAPS/PC–128 – Which BrainMakers uses
The Mosaic QED Board The HNC Balboa Board
Acceleration bards are extremely useful because they reduce training
time, which is usually quite long.
Intelligent Agents and Creativity
An intelligent Agent (IA) is a computer program that helps a user with
routine computer tasks.
The first vignette is an example of a first generation individually
operated agent.
They can overcome the most critical limitation of the internet – the
information overflow – and make electronic commerce a viable
organization tool.
Names
Several names are used to describe intelligent agents, including
software agnates, wizards, knowbots, and softbots (intelligent software
robots).
The names sometimes reflect the nature of the agent.
Characteristics of Intelligent Agents
Autonomy Learning
Operates in background Reactivity
Single task Proactive ness
Communication Temporal continuity
Automates repetitive tasks Personality
Supports conditional processing Mobility
Autonomy
An agent is autonomous; that is, it is capable of action on its own.
An agent must be able to make some decisions on its own by being
goal oriented, collaborative, and flexible.
It must b able to alter its course or behavior when it meets obstacles to
find ways around the impediment.
Example _ an agent should be able to accept high level request and
decide on its own where and how to carry out the request.
Autonomy implies that an agent takes initiative and exercises control
over its own actions in the following ways.
Goal oriented Accepts high level requests indicating what a human wants and
is responsible for deciding how and where to satisfy certain
requests.
Collaborative Does not blindly bet commands, but can modify requests, ask
clarification questions, or even refuse to satisfy certain requests.
Flexible Actions are not scripted, able to dynamically choose which
actions to invoke and it what sequence, in response to the state
of its external environment.
Self Starting Unlike standard programs directly invoked by user, an agent can
sense changes of its environment and decide when to act.
Operates in backboard
An agent must be able to work out of sight, within the realm of
cyberspace or other computer systems, without the constant attention
of its user.
Some developers use the term remote execution or mobile agents to
describe this attribute.
Single task
In most cases an agent is designed to accomplish a single task.
A single task cold is searching the internet for articles.
Another might be filtering electronic mail.
The more futuristic agents appear capable of doing multiple things.
Communication
Many agents are designed to interact with other agents, humans, or
software programs.
Instead of making a single agent smarter, additional agents can be
created to handle undefeated tasks.
The ability to collaborate enables the system to accomplish more
complex tasks.
Automates repetitive tasks
An agent is designed to do narrowly defined tasks, which it can do over
and over without getting bored or sick or going on strike.
No comments:
Post a Comment