Neural Net is a learning algorithm that is inspired by how the human brain works.
For example, imagine you love chocolate cake so much that, you joyfully exercise a bit more during the week just to enjoy that delicious chocolate cake without feeling guilty.
But if the weather is terrible there is no way you go exercise, and then you can’t eat the delicious chocolate cake. Although, if your beautiful girlfriend / boyfriend exercise with you, then you ignore the weather and joyfully exercise and then you can enjoy that delicious chocolate cake without feeling guilty.
The brain’s nervous system passes information using a synapse structure which allows neurons to pass information to other neurons and finally make a decision. This structure of passing information and decision making is the construction behind neural net algorithm.
The data structure provides weights on the edges for the nodes/synapses in a directed graph. For example, our chocolate cake decision making could be translated into:
w1=6 Whether or not your girlfriend or boyfriend exercise together with you
w2=3 Enjoying delicious chocolate cake
The high weights for example indicates the condition to have a high influence on your output decision making, while lower weight is not that influential.
The illustration below is an example of neural net with 5 different input of information:
The edges/arrows represent weights each input/node have a weight associated with it. These weights are applied when training neural net.
Three of the inputs could represent 1= Delicious chocolate cake, 2= The weather, 3= Your girlfriend or boyfriend, and the last two inputs could be other information.
The output is the condition of the decision determined by the hidden layer.
ThingWorx Analytics Server applies neural net with full interconnection layer, which means each value from the input layer is duplicated and sent to each node in the hidden layer, just like in following illustration.