Hi everyone! I hope all are you doing well. My Question is If we are solving a problem by using Neural Networks we need to define boundaries in the data? like in coffee roasting example, we define the over cook region, under cook region, best cook region in the data??
The ranges are defined by the data in the training set.
But think again - While using the Neural network, did you define the boundaries of each region anywhere?
What we had with us was the data. In a very simple case, we can plot the data and visually understand the boundaries and maybe even translate that understanding in the form of an equation that represents the boundary. But this becomes humanly impossible 1) as the number of dimensions (features) of the data increases and 2) as the complexity of the boundary grows.
The neural network (or any of the advanced models for that matter) will explore through this complexity and find the regions and boundaries which would best segregate the various classes within the data. Once it has understood the boundaries, it maintains this information in the values of the weights and biases.
So, the next time we ask the NN to predict for a given set of inputs, it will call upon its knowledge of the boundaries of each class which is stored in the weights and biases of the various neurons, and thereby provide us with the correct classification.
1 Like