2. Index
What is the Multi-layer perceptron
Why MLP
Architecture of MLP
Who its work
Application of MLP
Example
3. Multi-layer perceptron
MLP is a class of feedforward artificial neural networks. An
MLP consists of, at least, three layers of nodes: an input
layer, a hidden layer and an output layer. Except for the
input nodes, each node is a neuron that uses a nonlinear
activation function. MLP utilizes a supervised learning
technique called backpropagation for training.
MLP useful in research for their ability to solve problems
stochastically.
4.
x1
x2
Xn
A neuron can have any number of inputs from one to n, where n is the
total number of inputs.
The inputs may be represented therefore as x1, x2, x3… xn.
And the corresponding weights for the inputs as w1, w2, w3… wn
Output a = x1w1+x2w2+x3w3... +xnwn
process output
Activation
function
weight
w0
-1
+1
6. t/f
t/f
z
Sin(x)= 1 if x ≥ 0
0 if x < 0
OR T F
F
T
T F
TT
output
XOR T F
F
T
T F
TF
7. Why MLP
Single neurons are notable to solve complex tasks(e.g.
restricted to linear calculations).
Creating networks by hand is too expensive we want to
learn from data.
We want to have a generic model that can adapt to
some training data.
8. Architecture of MLP
Input layer
Hidden layer
Output layer
A multi layer perceptron's (MLP) is a finite acyclic graph.
The nodes are neurons with logistic activation.
Summation transformation
S=∑w.x
∫(s)= 1
1+e-s
9. Connection layers
• No direct connections between input and output layers.
• Fully connected between layers.
• Number of output units need not equal number of input
units.
• Number of hidden units per layer can be more or less than
input or output units.
10. Who its work
The input value are presented to the perceptron and if the
prediction output is the same as the desired output, then the
performance to the weights are made.
However, if the output doesn’t match the desired output the
weights need to be changed to reduce the error.
∆W=b*d*x
d: predicted output (desired output)
b: learning rate, usually less than 1 (beta time)
X: input data
11. Application of MLP
MLPs are useful in research for their ability to solve
problems stochastically, which often allows approximate
solutions for extremely complex problems like fitness
approximation.
In MLPs can be used to create mathematical models by
regression analysis.
MLPs make good classifier algorithms.
If we are assume that we have this canvas and we have a whole bunch of point in that canvas and we draw a line between them and we trying to classify some point that are one side of the line and the some other points that are only another side of line.
we can call it neuron or processor and receiving input it had from x0 and x1
Each one of these inputs was connected to the processor with the weight and the processor created a sum of all the inputs multiplied by the weight. That weight sum is passed through an activation function to generate the output.
The question here what is the limit here so the idea is that in different machine learning applications let’s take a very classic classification algorithms.
when we say, if we have a handwritten digits like numbers (8) and I have all the pixel of this digits and I want these pixel input to the perceptron and I want the output tell me a set of probabilities.
So the idea here that take a random number and put it in the input like 28*28=784 pixel image of grayscale values and those they are coming into processor which was wait to sudden and get the output
So, if I have a hole bunch more inputs and a hole bunch of outputs but still have a single processor unit, the reason that can came to published a book in 1969 by Marvin Minkey and Seymour that said the single perceptron can only solve the linearly separable problem.
So let’s think about this over here a linearly separable problem meaning I need to classify this stuff If I were to visualize all that stuff I can draw a line between stuff in this class and stuff in that class .
That’s mean is And and or is linearly separable problem
We found the or and and are separable to linearly perceptron
By assume that this node is AND can automatically give me the output And
But if we connected by another perceptron in OR connecting every node can give the connection of that node
So this perceptron cannot solve AND this perceptron can solve OR. The idea here are more complex problems that are not linearly separable can be solve by linked a multi layer perceptron .
In this case it should be gone more further step to Multi-layer perceptron
And talking about the logical gate AND OR and XOR
No direct connections between input and output layers.
Fully connected between layers.
Often more than 3 layers.
Number of output units need not equal number of input units.
Number of hidden units per layer can be more or less than input or output units.