Scribd
Upload
4 views3 pages

NNFL Module3 Supervised Learning Revision

The document provides a revision of supervised learning concepts, focusing on neural network architectures such as Single Layer Perceptron, Bipolar AND Perceptron, Adaline Network, and Multilayer Perceptron (MLP). It outlines key features, learning rules, and advantages of each model, as well as the error backpropagation algorithm used for training MLPs. Additionally, it includes common exam questions and a quick revision section summarizing the main points.

Uploaded by

radhasiyar124
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views3 pages

NNFL Module3 Supervised Learning Revision

The document provides a revision of supervised learning concepts, focusing on neural network architectures such as Single Layer Perceptron, Bipolar AND Perceptron, Adaline Network, and Multilayer Perceptron (MLP). It outlines key features, learning rules, and advantages of each model, as well as the error backpropagation algorithm used for training MLPs. Additionally, it includes common exam questions and a quick revision section summarizing the main points.

Uploaded by

radhasiyar124
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

NNFL – Module 3 Supervised Learning Revision

1. Single Layer Perceptron

Definition: A perceptron is a single layer feedforward neural network used for classification of linearly
separable data.

Features: Supervised learning, step activation function, binary classifier, linear problems only.

Cannot solve XOR because XOR is not linearly separable.

Perceptron Learning Rule: w(new)=w(old)+η(t−y)x

Bias Update: b(new)=b(old)+η(t−y)

Learning Rate η controls speed of learning.

If error = 0, no weight or bias update occurs.

Exam Tip: Always write net, output, error, weight update and bias update clearly.

2. Bipolar AND Perceptron

Binary to Bipolar Conversion: 0→-1 and 1→+1

Activation Function: Output = +1 if net > 0 else -1

Important Point: Error may become 2 because 1 - (-1) = 2.

Best Exam Format: Use tabular format with x1, x2, target, net, output, error, ∆w1, ∆w2 and ∆b.

3. Adaline Network

Definition: Adaline (Adaptive Linear Neuron) uses linear activation function and LMS/Delta learning
rule.

Most Important Difference: Perceptron learns using output y whereas Adaline learns using net value.

Activation Function: f(x)=x

LMS Rule: w(new)=w(old)+η(t−net)x

Bias Update: b(new)=b(old)+η(t−net)

Advantages: Stable learning, minimizes mean square error.

Common Mistake: Using t−y instead of t−net.


4. Multilayer Perceptron (MLP)

Definition: MLP is a feedforward neural network with one or more hidden layers.

Need of Hidden Layer: Required for solving XOR and non-linear problems.

Structure: Input Layer → Hidden Layer → Output Layer

Activation Functions Used: Sigmoid, Tanh, ReLU

Advantages: Better accuracy and complex pattern learning.

Limitation: More computation and training time.

Most Important Line: MLP solves XOR problem.

5. Error Backpropagation Algorithm

Definition: Backpropagation is a supervised learning algorithm used for training MLP by propagating
error backward.

Core Idea: Forward pass generates output, backward pass corrects weights.

Steps: Initialize weights → Forward propagation → Error calculation → Backward propagation →


Weight update → Repeat.

Weight Update Formula: w(new)=w(old)+ηδx

Sigmoid is used because it is differentiable.

Advantages: Solves non-linear problems and improves accuracy.

Limitations: Slow training and computationally expensive.

Memory Trick: Forward for output, backward for correction.

6. Most Expected University Questions

• Explain perceptron learning algorithm

• Implement AND gate using perceptron

• Difference between Perceptron and Adaline

• Explain Adaline with LMS rule

• Explain MLP architecture

• Why hidden layer is needed?


• Explain backpropagation algorithm

• Why perceptron cannot solve XOR?

7. Ultra Quick Revision

Perceptron → Linear classifier

Adaline → Learns using net

MLP → Uses hidden layers

Backpropagation → Trains MLP

XOR → Non-linearly separable

You might also like