The artificial neural network which is simply called neural networks is an artificial interconnections of group of nodes that try’s as much as possible to mimic the biological brain, it is made up of three types of layers, input, hidden and output layers, as seen in the diagram above. the input layers receive inputs then pass it to the hidden layers, the hidden layer can be more than one depending on the goals and object of we want to archive the process of passing input from the input layer to the hidden layer is called forward propagation, the hidden layer process this input with the help of sophisticated mathematical method and try to predict a targeted output (output layers), the error in prediction is taken into consideration which is called weight, the neuron keeps adjusting the weight by sending data backward from on layer to another and making adjustment this process is called backward propagations. With forward and backward propagations the neuron keep learning from the data and adjusting the weight until a near perfect prediction is made.
There are many types of Artificial Neural Network being used in Machine learning but the two famous are CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network)
CNN Mainly used in image recognitions CNN uses sophisticated mathematical method called Convolution, convolution is an operation in mathematics where two functions combine to produce a third function that expresses how the shape of one affect the other.
RNN RNN stands for Recurrent Neural network and it is a type of network in deep learning Neural Network specifically design for sequential data such as stock trading, forest trading, Speech recognition, Natural Language Processing like speech and handwriting recognition. RNN has memory, when we propagate forward or backward we keep the information from the hidden layer in other words the hidden layers cannot communicate among themselves. But with RNN hidden layers can communicate together and keep information of their previous inputs. RNN is a very deep neural Network and because of the interconnection of its hidden layer it need high processor to compute.
Check out this deep learning project in which I used LSTM a part of RNN to analyze and predict stock price data