Home

Welcome to the deeplearning4kids wiki!

Whats an Artificial Neuron

Its a function which takes many inputs and generate output if the inputs are good enough to trigger the threshold function.

Whats a threshold function inside a neuron

Any function which could generate values through 0 to 1 or generate binary values 1 or 0 through out the input domain is called a threshold function. Following are some examples

1. The simplest function is that outputs 0 when its inputs are below a certain threshold, and outputs 1 when its inputs are above that threshold. The major drawback with this function is that it doesn't have ""region of uncertainty"", which is not good when u want neurons to learn.
2. The other set of functions are of type which are continuous in nature, eg logistic sigmoid, tanh or hinge loss function

How to train ur dragon I mean Neuron##

1. Give input to neuron
2. Let it produce an output
3. Check how far the result is with reality
4. Penalise Neuron based upon the difference between result and reality
5. Repeat the whole process until result is in acceptable limits

Whats a Neural net

When many neurons are connected, they form a neural net, what could be achieved depends upon how they are connected and how many of then are there. Generally many neuron’s output goes as input to next level neuron. i.e. The input to a neuron's inner function, denoted by x, is a sum of other neurons' outputs. To bring in flexibility instead of simple sum we will do weighted sum with basis, hence its input value multiplied by weight and a bias value, which means even if the actual output value from on of the feeder neuron is 0, the bias will contribute. Mathematically speaking weight controls the scaling of the sigmoid function, while basis value translates the function left and right.

What are layers in neural network

A basic neural network consists of three layers, called the input layer, hidden layer, and output layer. Each layer consists of one or more nodes, and information flows from input layer to hidden layer and finally to output layer. Generally input layer are passive, meaning they do not modify the data. They receive a single value on their input, and duplicate the value to their multiple outputs.

How do you train the network

Lets assume we want our network to classify a hand-drawn digit, which means if we provide our network with a hand-drawn image of digit 0, the output should be [1,0,0,0,0,0,0,0,0,0], similarly if the provide an image of digit 9, the output should be [0,0,0,0,0,0,0,0,0,1]. Which also kind of imply that our output layer will have only 10 neurons.