Transcript Neural Nets Applications -

Neural Nets
Applications
Introduction
Outline(1/2)
1. What is a Neural Network?
2. Benefit of Neural Networks
3. Structural Levels of Organization
in the Brain
4. Models of a Neuron
5. Network Architectures
6. Artificial Intelligence and Neural
Networks
Outline(2/2)
7. Existing Applications
8. Possible Applications
9. Experiment I
10. Experiment II
11. Other names for Neural Networks
12. Who are the key player?
What is a Neural
Networks(1/5)
 Neural networks technology is not
trying to produce biological machine
 but is trying to mimic nature’s
approach in order to mimic some of
nature’s capabilities.
What is a Neural Networks
(2/5)
Definition:
A neural network is a massively parallel
distributed processor that has a natural
propensity for storing experiential
knowledge and making it available for
use.
What is a Neural Networks
(3/5)
It resembles the brain in two respects:
1.
2.
Knowledge is acquired by the network
through a learning process.
Interneuron connection strengths
known as synaptic weight are used to
store the knowledge.
What is a Neural Networks
(4/5)

The Human Brain:




Five to six orders of magnitude slower than
silicon logic gates
With 60 trillion synapses or connections
A highly complex, nonlinear, and parallel
computer.
Figure 1.1
What is a Neural Networks
(5/5)
Benefits of Neural Networks
(1/2)
1.
2.
3.
4.
5.
Nonlinearity
Input-Output Mapping
Adaptivity
Evidential Response
Contextual Information
Benefits of Neural Networks
(2/2)
Fault Tolerance
7. Implementability
8. Uniformity of Analysis and Design
9. Neurobiological Analogy
6.
Structural Levels of
Organization in the Brain (1/3)
Figure 1.2
2. Figure 1.3
1.
Structural Levels of
Organization in the Brain (2/3)
Structural Levels of
Organization in the Brain (3/3)
Models of a Neuron (1/6)
Figure 1.4
Three basic elements of the neuron model:
1.
2.



A set of synapses or connecting links, each of
which is characterized by a weight or
strength of its own.
An adder for summing the input signals,
weighted by the respective synapses of the
neuron; the operations described here
constitute a linear combiner.
An activation function for limiting the
amplitude of the output of a neuron.
Models of a Neuron (2/6)
Models of a Neuron (3/6)
3.
Mathematical terms:
p
where:
xj: input signals
wkj: synaptic weights
uk: linear combiner
output
θk:: threshold
f() : activation function
yk: output signal
u k   wkj x j
j 1
and
y k  f (u k   k )
Models of a Neuron (4/6)
4. Types of activation function:
a. Threshold function
Models of a Neuron (5/6)
4. Types of activation function:
b. Piecewise-linear function
Models of a Neuron (6/6)
4. Types of activation function:
c. Sigmoid Function
Network Architecture (1/5)
1. single-layer feedforward network
Network Architecture (2/5)
2. Multilayer feedforward network (fully
connected
Network Architecture (3/5)
2. Multilayer feedforward network (partially
connected
Network Architecture (4/5)
3. Recurrent networks
Network Architecture (5/5)
4. Lattice Structures
Artificial Intelligence and
Neural Networks (1/5)
AI system
Artificial Intelligence and
Neural Networks (2/5)
a. Representation
- use a language of symbol structures to
represent both general knowledge about a
problem domain of interest and specific
knowledge about the solution to the
problem.
Artificial Intelligence and
Neural Networks (3/5)
b. Reasoning
- the ability to solve the problems
- be able to express and solve a broad range
of problems and problem types.
- be able to make explicit and inplicit
information known to it
- have a control mechanism that determines
which operations to apply to a particular
problem.
Artificial Intelligence and
Neural Networks (4/5)
c. Learning
- Fig 1.27
- Inductive, rules are from raw data and
experience
- Deductive, rules are used to determine
specific facts
Artificial Intelligence and
Neural Networks (5/5)
Existing Applications(1/4)
1. Long distance echo adaptive fitter
adaptive noise canceling
-- ADALINE
2. Mortgage risk evaluator
3. Bomb sniffer
-- SNOOPE
-- JFK airport
Existing Applications(2/4)
4. Process Monitor
-- GTE used in fluorescent bulb
plant.
-- To determine optimum manufacturing
condition.
-- To indicate what controls need to be
adjusted , and potentially to even shut
down the line.
-- Statistics could provide same result but
with huge data.
Existing Applications(3/4)
5. Word Recognizer
--Intel used single speaker on limited
vocabulary.
6. Blower Motor Checker
--Siemens used to check Blower motor
noise is heater.
7. Medical events
Existing Applications(4/4)
8. US postal office for hand written
9. Airline marketing tactician.
Possible Applications(1/6)
1. Biological
--Learning more about the brain
and other systems
--Modeling retina , cochlea
2. Environmental
--Analyzing trends and patterns
--Forecasting weather
Possible Applications(2/6)
3. Business
--Evaluating probability of oil in
geological formations
--Identifying corporate candidates for
special positions
--Mining corporate databases
--Optimizing airline seating and fee
schedules
--Recognizing handwritten
characters, such as Kanji
Possible Applications(3/6)
4. Financial
--Assessing credit risk
--Identifying forgeries
--Interpreting handwritten forms
--Rating investments and analyzing
portfolios
Possible Applications(4/6)
5. Manufacturing
--Automating robots and control
system (with machine vision and
sensors for pressure. temperature, gas,
etc.)
--Controlling production line processes
--Inspecting for quality
--Selecting parts on an assembly line
Possible Applications(5/6)
6. Medical
--Analyzing speech in hearing aids for
the profoundly deaf
--Diagnosing/prescribing treatments from
symptoms
--Monitoring surgery
--Predicting adverse drug reactions
--Reading X-rays
--Understanding cause of epileptic
seizures
Possible Applications(6/6)
7. Military
--Classifying radar signals
--Creating smart weapons
--Doing reconnaissance
--Optimizing use of scarce
resources
--Recognizing and tracking targets
Experiment I
1. to understand a sentence are
2.
3.
4.
5.
character a time is much larger than
one word a time
conventional computer processes
its input one of a time, working
sequentially
our eyes look at the whole sentence
vowels are missing
three different groupings
Experiment II(1/2)
1. Toss a chalk to another one
-- it is hard in dynamics
-- estimate the speed , the
trajectory, the weight
-- in real time
-- computer must be faster
Experiment II(2/2)
But
-- our brain is lower than
computer
-- our brain still better than
computer
Why?
 parallel processing
Other Names for Artificial Neural
Networks
Parallel/distributed processing models
Connectivist/connectionism models
adaptive systems
self-organizing systems
Neurocomputing
Neuromorphic systems
Self-learning systems
Who Are the Key Players?
(1/2)
1. Medical and theoretical neurobiologists
--Neurophysiology, drug chemistry ,
molecular biology
2. Computer and information scientists
--Information theory
3. Adaptive control theorists/psychologists
--Merging learning and control theory
Who Are the Key Players?
(2/2)
4. Adaptive systems
-- researchers/biologists
--Self-organization of living species
5. AI researchers
--Machine learning mechanisms