Unit 1

Soft computing
it is used off approximate calculations to provide imprecise but usable solutions of complex
computational problems. It provides an approach if problem solving means other than computers.
With the human mind as the role model, soft computing is tolerance partial truths, uncertainty,
imprecision and approximation.

soft computing uses fields of study in

1. fuzzy logic
2. artificial neural network
3. machine learning
4. expert system
5. genetic algorithm

Difference between soft computing and hard computing

Applications of soft computing

1. It is widely used in gaming products like poker etc.

2. In kitchen appliances life such as microwave etc.
3. In most used home appliances like washing machine, heater etc.
4. Apart from all these uses, it is also used in robotics work.
5. Image processing and data compression are also popular application of soft computing.
By : Neha Tyagi
 6. Used for handwriting recognition.

Need of soft computing

1. Hard computing is used for solving mathematical problems that need a precise answer. It
failed to provide solutions for some real life problems. Therefore for real life problems whose
precise solution does not exist, soft computing helps.
2. When conventional mathematical an analytical model fails, soft computing helps.
3. Soft computing is not limited to theory, it also insights into real life problems.
4. Soft computer helps with real life problems which does not have ideal case.

Neuron

Fig of neuron

A nerve cell Is a special biological cell that process in formation. It is a small cell that receives
electrochemical signal from its various sources and transmit electrical impulse to other neurons

A neural network is a series of algorithms that help to recognise underline relationship in a set of
data through a process that mimics the way the human brain operates.

structure of a neuron

1. Dendrites They are tree like branches , responsible for receiving the Information from
other neurons and they are connected with each other too.
2. Soma It is the cell Space of neurons and is responsible for process of information, they
have received from dendrites
3. Axon it is just like a cable through which neurons send the information
4. Synapses it is connection between axon and other neurons dendrite.

Artificial Neural Network ( ANN )

Artificial neural network is efficient computing system who's control key is borrowed from
biological neural network full stop it is a computational based biological neural network full
stop artificial neural networks are also named as artificial neural system or parallel distributed
processing system for it

Borrowed from biological neural network. It is a computational based biological neural

network. Artificial neural network are also named as artificial neural system or parallel
distributed processing system for it stop artificial neural network acquires the large collection

By : Neha Tyagi
 of units that are interconnected in some pattern to allow communication between units, also
called neurons or nodes.

Different terminology used in biological neural network an artificial neural network-

BNN ANN
Soma Neuron / Node
Dendrites Input
Synapses Weight or
interconnections
Axon Output

Difference between biological neuron an artificial neuron

Characteristics of artificial neural network

• Newly implemented mathematical model

• Huge number of interconnected processing elements called neurons.
• Information is stored in neurons are basically the weighted linkage of neurons.
• Input signal arrives at processing elements through connections and connecting weights.
• It has ability to learn, recall and generalise from given data by suitable assignment and
adjustment of weights.
• Collective behaviour of neurons describe its computational power and single neuron
carries specific information.

By : Neha Tyagi
Architecture of artificial neural network

This architecture or model Is divided is divided into 3 layers (also called elements of neural
network)
1. Input layer
• It contains those notes or artificial neurons that receive input from outside word on
which network will learn, recognise and understand them or process otherwise.
• No computation is performed in this layer
2. Output layer
• it contain notes that respond to information about how it is learnt any tasks
3. Hidden layer
• These are in between input and output layer line change.
• This layer perform all sort of computation on the features and 3 rd through the input
layer and transfer the result to the output layer

key points-

• There can be 0 or more hidden layers in artificial neural network.

• No hidden layers depends on complexity of problem.
• 1 hidden layer is sufficient for large majority of problems.
• These are always be an input and output layer.

Activation function (AF)

By : Neha Tyagi
 • Activation function defines the output of neuron in a range of values.
• It defines whether a neuron should be activated or not.
• The normalised amplitude range of output of neuron is written as closed unit interval [0, 1]
or [-1, 1].
(Diagram same as model of ANN)

Uses of activation function

1. Help neuron models in learning.

2. Help in mapping from input to output
3. Main purpose is to convert input signal to output signal.

Necessity of activation function

• Helps to calculate exact output of A and N

• introduce non-linear properties of a word network.

What would happen if we don’t use activation function?

Without our activation function, output signal would be work as simple linear function, which might
be easy to solve but limited in processing complex learning. This is because in simple linear function
we use polynomial of 1 degree only. But if we deal with images, videos, audio than simple linear
function will fail.

Types of Activation function

By : Neha Tyagi
By : Neha Tyagi
By : Neha Tyagi
What do you understand by the term “Recurrent Network “?

Then outputs can be directed back as inputs to the same layer or proceeding layer notes, then it
results in feedback networks. Recurrent networks are feedback networks with closed loops.

Learning methods

• it refers to the processor in which learning rules are used for adjusting the weights.
• In artificial neural network, learning methods can be viewed as the problem of updating
network architecture and connection weights so that our network can efficiently perform a
task.

By : Neha Tyagi
 These three types are classified based on-

• presence or absence of teacher

• information provided for system to learn.

Supervised learning

• Learning is performed with the help of a trainer.

• In artificial neural network, each input vector requires a corresponding target vector, which
represent the desired output.
• The input vector along with the target vector is called training pair.
• The input vector result in output vector.
• the actual output vector is compared with the desired output Vector.
• If there is a difference means an error signal is generated by the network. It is used for
adjustment of weights until actual output machines desired output comes.

Unsupervised learning

• Learning is done without the assistance of a trainer.

• During the training process, the network accept input patterns and uninjured them into
clusters. There is no feedback from the environment to inform what the output should be or
whether it is correct.
• The network discovers patterns, regularity’s, features/ categories and relationship further input
data over the output.

By : Neha Tyagi
 • Precise clusters are generated by recognising similarities and differences, a process known as
self organisation.

Reinforcement learning

• A teacher is present but does not present the expected or desired output but only indicated if
the computed output is correct or incorrect.
• If the computer output is correct, then it will get a positive feedback (reward )or if the
computer output is incorrect then a negative feedback (penalty) is received.
• In this type of learning the artificial neural network learns itself automatically using the
feedback either positive or negative.
• Since there is no expected output so the artificial neural network is bounded to learn by
experience only.
• Reinforcement learning solves specific type of problems where decision-making is sequential
and the goal is long term.
• The primary goal of artificially neutral network is to improve the performance by getting the
maximum positive rewards.
Key feature
1. It is based on the hit and trial process.
2. Annan takes the next action and changes parameters according to the feedback of the
previous section.
3. In this there is no instruction about the environment and what action needs to be taken.

By : Neha Tyagi
By : Neha Tyagi
By : Neha Tyagi
By : Neha Tyagi
By : Neha Tyagi