Multilayer neural, Back propagation network and radial basis function network
Multilayer Neural Network (MLNN)
A neural network is a computer model that works like the human brain.
It takes input.
It processes the input.
It gives output.
A multilayer neural network has:
One input layer
One or more hidden layers
One output layer
Hidden layer means a middle layer that helps in better learning.
Simple Example
Input: Marks in Maths and English
Hidden layer: Combines marks and finds pattern
Output: Pass or Fail
Like a teacher who checks many things before giving final grade.
Structure of Multilayer Network
It has three main parts:
-
Input Layer
Takes data
Example: Student marks
-
Hidden Layer
Does calculation
Finds patterns
-
Output Layer
Gives result
Example: Grade A, B, C
Simple Diagram (Text Form)
Input → Hidden Layer → Output
Why We Need Multilayer Network?
Single layer network cannot solve complex problems.
Multilayer network:
Solves complex problems
Finds deep patterns
Gives better accuracy
Example
Single layer: Can check only pass/fail
Multilayer: Can predict exact grade
Like:
One teacher checks homework quickly
Panel of teachers checks deeply
Comparison: Single Layer vs Multilayer Network
| Feature | Single Layer | Multilayer |
|---|---|---|
| Hidden layer | No | Yes |
| Handles simple problems | Yes | Yes |
| Handles complex problems | No | Yes |
| Accuracy | Low | High |
| Learning ability | Limited | Better |
Real-Life Example
Single layer = One security guard checking ID
Multilayer = Guard + Biometric + CCTV
Exam Tip
If question asks:
“Why multilayer network is better?”
Write:
Has hidden layers
Solves complex problems
Gives better accuracy
Back Propagation Network (BPN)
Introduction
Back Propagation is a learning method used in multilayer networks.
It helps the network:
Find mistakes
Correct mistakes
Improve output
Back propagation means:
Send error back
Adjust weights
Weight means importance of input.
Architecture of Back Propagation Network
It has:
Input layer
Hidden layer
Output layer
Weights between layers
Input → Hidden → Output
↑
Error comes back
Back Propagation Algorithm (Step-by-Step)
Step 1: Give input
Step 2: Calculate output
Step 3: Compare with
actual output
Step 4: Find error
Step 5: Send error backward
Step
6: Adjust weights
Step 7: Repeat until error becomes small
Simple Example
Think of exam preparation:
You attempt test.
Teacher checks mistakes.
You correct mistakes.
You improve next time.
That is back propagation.
Local Minima and Global Minima
When training network:
Network tries to reduce error.
Error surface looks like hills and valleys.
Global Minima
Lowest possible error
Best solution
Local Minima
Small valley
Not best but looks good nearby
Real-Life Example
Imagine trekking:
Local minima = Small valley between hills
Global minima = Deepest valley
Network sometimes stops at small valley.
Exam Tip
Define:
Local minima: Small lowest point
Global minima: Absolute lowest point
How to Improve Back Propagation Performance
Heuristics (simple tricks):
-
Use proper learning rate- Learning rate = speed of learning
-
Normalize data - Make data in same range
Use good initial weights
Use more training data
Example
If you learn too fast:
You miss concepts.
If you learn too slow:
You waste time.
Same with learning rate.
Applications of Back Propagation
Image recognition
Handwriting recognition
Face detection
Medical diagnosis
Stock prediction
Real-Life Example
Face unlock in mobile phone
Google photo tagging
Radial Basis Function (RBF) Network
Introduction
RBF network is another type of neural network.
It also has:
Input layer
Hidden layer
Output layer
But hidden layer works differently.
Architecture of RBF Network
Structure:
Input → RBF Hidden Layer → Output
Hidden layer uses special function called:
Radial Basis Function
(It
measures distance)
Distance means how close input is to known data.
Simple Example
In shopping:
If product looks similar to what you like,
System recommends it.
RBF checks similarity.
Training Algorithm of RBF
Training happens in two steps:
Step 1:
Choose centers (important points)
Step 2:
Adjust output weights
It trains faster than back propagation.
Approximation Property of RBF
Approximation means:
It can copy or match any complex function.
RBF can:
Model complex patterns
Give smooth output
Example
Weather prediction:
Temperature changes smoothly.
RBF gives smooth curve.
Comparison: RBF vs Back Propagation
| Feature | Back Propagation | RBF |
|---|---|---|
| Training speed | Slow | Fast |
| Complexity | More | Less |
| Hidden layer | Multiple possible | Usually single |
| Accuracy | High | High |
| Local minima problem | Yes | Less |
Real-Life Example
Back Propagation:
Like preparing full semester notes.
RBF:
Like short revision notes.
Both useful, but RBF is faster.
Important Exam Questions
Short Questions
Define multilayer neural network.
What is back propagation?
Define local minima and global minima.
What is RBF network?
Write two applications of BPN.
Long Questions
Explain back propagation algorithm with steps.
Compare single layer and multilayer network.
Compare RBF and Back Propagation networks.
Explain architecture of RBF network.
Remember This
Multilayer = Has hidden layer.
Back propagation = Error correction method.
Local minima = Small valley.
Global minima = Deep valley.
RBF = Distance-based network.
Quick Revision Table
| Topic | Key Idea |
|---|---|
| Multilayer Network | Has hidden layers |
| Back Propagation | Error goes backward |
| Local Minima | Small error point |
| Global Minima | Lowest error point |
| RBF | Similarity based learning |
Final Summary (Quick Revision)
Multilayer networks solve complex problems.
Back propagation improves network by correcting errors.
Local minima can trap network.
RBF network learns faster.
Both BPN and RBF are important for pattern recognition.
Focus on:
Architecture
Algorithm steps
Comparison tables
Applications
These points help you score well in exams.