Sie sind auf Seite 1von 38

CHAPTER 2

NEURAL NETWORK FOR CONTROL


Outline
 Neural Network for Control:
General principle, biological neuron, artificial
neuron, network structure, activation equation

Learning Outcome
 Able to understand the principle of neural network

ASNOR/KEE/UPM/EEE4404
Main events in the history of AI
What is Artificial Neural Network
(ANN)?
• ‘Artificial neural network’ or ANN is the term used to
describe a computer model assumption of the biological
brain.
• It consists of a set of interconnected simple processing
units which combine to output a signal to solve a certain
problem based on the input signals it received.
• The interconnected simple processing units have
adjustable gains that are slowly adjusted through
iterations influenced by the input output patterns given to
the ANN.
What is Artificial Neural Network (ANN)?
(cont)
• An ANN is an information-processing system that has
certain performance characteristics in common with
biological neural networks.
• Basically it is a system that handles many input signals,
processes them, and output them to solve a task that it
has been trained to solve.
What is Artificial Neural Network (ANN)?
(cont)
• ANNs have been developed as generalizations of mathematical
models of human cognition or neural biology, based on the
assumptions that:
 An information processing system, developed based on the biological
brain.
 Consists of simple processing units called neurons (nodes).
 All the neurons are assumed to be interconnected.
 Neurons are activated by a certain (activation) function.
 For each interconnection, there is an associated weight (adjustable
gains).
 Signals are processed from input to output by multiplying the weights
and activated signals.
 A learning rule is needed to adapt the weights to solve a particular
problem.
 Training patterns are needed in order to train the ANN.
Some Keywords related to ANN
• Neuro-models
• Learning / Teaching Phase
• Activation functions
• Model parameters
• Training Data
• Inputs / Outputs
• Error
• Weights
• Adaptation
• Test / Simulation
• Application
Application of ANN

• ANNs are learning systems that can be applied into


many types and areas of applications.
• Basically an ANN maps one function into another
and they can be applied to perform the following:
 Pattern recognition
 Pattern matching
 Pattern classification
 Prediction
 Clustering
 Control
 etc
Some examples of areas of applications of ANN
• Financial Recognizing Genes
Stock Market Prediction Biological Systems Analysis
Bankruptcy Prediction
Price Forecasts • Operational Analysis
Economic Indicator Forecasts Retail Inventories Optimization
• Medical Scheduling Optimization
Medical Diagnosis Managerial Decision Making
Detection and Evaluation of Medical Cash Flow Forecasting
Phenomena • HR Management
Treatment Cost Estimation
Employee Selection and Hiring
• Industrial Employee Retention
Process Control Staff Scheduling
Quality Control Personnel Profiling
Temperature and Force Prediction
• Energy
• Science Electrical Load Forecasting
Pattern Recognition Energy Demand Forecasting
Formulation Optimization Short and Long-Term Load Estimation
Identification Predicting Gas/Coal Index Prices
Physical System Modeling Power Control Systems
Ecosystem Evaluation Hydro Dam Monitoring
Polymer Identification
Basic Structure of a Neuron
From Biological Neuron to Artificial Neuron

All the inputs to the neuron would be summed up and


pass to an activation function to activate the neuron.
4 Main Components of a Neuron
• Soma: this is also called the cell body which supports
functions in the neuron.
• Axon: this is the part of the neuron
which carries the fired signal
out of the neuron.
• Dendrites: these are very dense
fiber-type of structure that are designed to receive
incoming signals.
• Synapses: these are the primary gateways where
neurons communicate with each other. There is
evidence that the chemical reactions at these synapses
are altered when learning takes place.
Analogy between biological and artificial
neural networks
Biological Neural Network Artificial Neural Network
Soma Neuron
Dendrite Input
Axon Output
Synapse Weight
A General Framework For Neuro-
Models
• A set of processing units or neurons
• A set of activation
• An output function for each unit
• A pattern of connectivity
• A propagation rule for propagating patterns of activities through the
network of connectivities
• An activation rule for combining the inputs impinging on a unit with
the current state of that unit to produce a new level of activation for
the unit
• A learning rule whereby patterns of connectivity are modified by
experience
• An environment within which the system must operate
A Set of Processing Units (Neurons)
• Also called neurons or nodes.
• Represent elements over which meaningful patterns
may be defined
• Need to define the role of each unit.
• Generally, we can characterize 3 types of units:
– input, hidden and output
• The units are labeled according to the location of their
layers.
Architecture of a typical artificial neural network

Output Signals
Input Signals

Middle Layer
(Hidden Layer)
Input Layer Output Layer
The neuron as a simple computing element
Diagram of a neuron
Input Signals Weights Output Signals

x1
Y
w1
x2
w2
Neuron Y Y

wn Y
xn
The State of Activation
• Is the activity of the processing element (neuron).
• Usually the state of activation of neuron can either be ON
(fired or 1) or OFF (unfired or 0).
• Is dependent on the application or task to be solved.
• Activation values can have minimum and maximum levels.
• Activation signals can be continuous or discrete.
• Typical examples of activation signals are:
 binary (0,1)
 bipolar (-1,1)
• Example of a sigmoid activation function
with continuous binary value
Output of Processing Units
(PU)
• Usually every neuron or PU has an output and the output
is a function of the input.
• The degree of activation of units determines the strength
of signal transmitted out.
• An Output function can have the following expression:

• Generally, in many models,


output level = activation level
A Pattern of Connectivity
• Every neural network must have a certain pattern of
connectivity.
• Simplest is to have the first layer neurons to be fully
connected to the second layer neurons (usually hidden
layer) and to have the second layer neurons to be fully
connected to the next layer (usually output layer).
• Some networks have feedback connections (recurrent
networks).
• The pattern of connectivity can be expressed in the form
of mathematical equations.
Example of Pattern of Connectivity
Rule of Propagation
• The flow of signals from the output units of a lower layer
to the input of units in an upper layer must follow a
certain rule which is called a rule of propagation.
• Such rule combines the output values with the
connectivity matrices.
• A typical example is given below:
Activation Rule

• The state of activation of the neuron tells whether the neuron


is fired (switched ON) or unfired (switched OFF).
• Thus, a function (aj) is needed to produce the new state of
activation which is called an activation rule.
• In general
• However, in many models

where the activation is the same as the output of the neuron.


• There are a number of activation functions commonly used
such as the logistic activation function, hyperbolic tangent
function, etc.
Examples of commonly used activation
function
A Learning Rule

• To modify the patterns of connectivity a learning rule is


needed as a function of experience.
• There are 3 ways of modifications :
 development of new connections
 loss of existing connections
 modification of the strengths of connections (or
weights) that already exist.
• The third way is the standard approach that is used in
many neural paradigms - major research area in neural
networks.
A Learning Rule (cont)

• Basically, the strengths of the connection weights (W)


are modified through an adaptation algorithm called
learning algorithm or rule.

• Thus W(new) = W(old) + W(new)


Representation of the Environment
• The neural network itself does not operate by itself to solve its own
problems.
• It is actually a tool used for solving a certain task and thus it must sit
in a certain environment.
• Typical applications are pattern recognition, pattern matching,
forecasting, control, etc.
• The environment may be characterized as a stable probability
distribution over the set of possible input patterns.
• Usually the set/type of input patterns determines the neural network
environment.
• Example of a neural network configuration for controlling a process
(as a controller).
General Information on ANN

• a) Biases and Thresholds


• b) Linear Separability
a) Biases and Thresholds
• A bias acts exactly like a weight.
• It is considered as a connection and its activation is always 1.
• It is then adapted similarly to the way a weight is adapted
according to the learning rule of the ANN.
• Its use is to increase signal levels in the ANN such as to
improve convergence.
• Some ANN do not use any bias signals.
• A neural network input S with a bias signal can be written as
follows: S = b +  wixi
Biases and Thresholds (cont)

• A threshold (θ) is a value that is used to make some form of


decisions in an ANN such that the ANN will fire or unfire.
• It is quite similar to a bias but is not adapted.
• An example of a binary threshold is given as follows:

• Then the equation of the separating line becomes:


b+w1x1+w2x2 = θ
b) Linear Separability

• Weights and biases in an ANN determine the


boundary regions that separate the ANN output
ANN Model

• Generally, an ANN is categorized by:


 Its pattern of connections between the neurons (also
called its architecture or model).
 the activation function used in the neurons.
 Its learning algorithm (which is the method of
determining its weights).
ANN Model (cont)

• Usually ANNs can be categorized into 2 models:

Feedforward
• All signals flow in one direction only, i.e. from lower
layers (input) to upper layers (output)

Feedback
• Signals from neurons in upper layers are fed back to
either its own or to neurons in lower layers.
Learning in ANN
• In all of the neural paradigms, the application of an ANN
involves two phases:
1. Learning phase
2. Recall phase

• In the learning phase (usually offline) the ANN is trained until


it has learned its tasks (through the adaptation of its weights)
while the recall phase is used to solve the task.
• An ANN solves a task when its weights are adapted through a
learning phase.
• All neural networks have to be trained before they can be
used.
Learning in ANN (cont)
• They are given training patterns and their weights are
adjusted iteratively until an error function is minimized.
• Once the ANN has been trained no more training is
needed.
• Two types of learning prevailed in ANNs:
1. Supervised learning:- learning with teacher signals or
targets
2. Unsupervised learning:- learning without the use of
teacher signals
Exercises
1. How does an artificial neural network
‘learn’?
2. Can a single neuron learn a task?
3. Why can a perceptron learn only
linearly separable functions?
4. Why do we need a hidden layer?

Das könnte Ihnen auch gefallen