Classifying Numbers As Odd Or Even With Neural Networks
5
1
110b (6)
isEven?
0
101b (5)
Neurons
Array => Number
3
-2
2
4
3
1
5
4
6
-1
0
1
Neurons
Input Layer: NOT Neurons
Neurons
Array => Number
???
5
2
* -1
* 4
-5
8
3
3
function Neuron(weights){
this.weights = weights
}
Neuron.prototype.process = function(inputs){
var sum = 0
for (var i=0; i<inputs.length; i++) {
sum += inputs[i] * this.weights[i]
}
return sum
}var neuron = new Neuron([4, -2, 6])
var output = neuron.process([1, 0, 1])
// output = 100
1
2
2.5
1
0
-3
1.5
1
0
1.5
Is 01b (1) even?
YES!
1.5 > 0.5
Need to adjust weights to get a more correct response!
Training Data
Number => isEven [0, 0, 0] => [1] [0, 0, 1] => [0] [0, 1, 0] => [1] [0, 1, 1] => [0] [1, 0, 0] => [1] ...
Hidden
Layer
5
9
4
0
2
generateTestData(0, 4)
[
{
"input": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
"output": [1]
},
{
"input": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1],
"output": [0]
},
{
"input": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0],
"output": [1]
},
{
"input": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1],
"output": [0]
}
]0000 0000 0000 0011b === 3
false / not even
var trainingSet = generateTestData(0, 100);
var testSet = generateTestData(10000, 11000);correctness([w1, w2, ...], tE) = xx% correctness([0.2, 2.9, ...], tE) = 60%
predictIsEven(number, weights) ==> e.g. 0.17 ==> 0.17 < 0.5, prediction is "not even"
correctness([1.2, 1.5, ...], tE) = 74% correctness([2.1, 0.3, ...], tE) = 23% correctness([1.3, 0.4, ...], tE) = 54%
exampleIsPredictedCorrectly(example, weights)
==> e.g. true
correctness(weights, trainingExamples)
==> e.g. 60%
CODE
Figure out what kind of changes to make
Make small random weight changes
Improvements
Backpropagation
0
1
2
2.5
1
0
-3
1.5
1
0
1.5
Is 01b (1) even?
YES
1.5 > 0.5
1
.7
1
1 > 0.5
0.7
0.7
0.3 < 0.5
NO
0.6
.5
0.3
0.7 > 0.5
0.5
Using neural nets to recognize handwritten digits
Slides
Full Code
Classifying numbers as odd or even with neural networks
By Matt Zeunert
