Neural Networks : Intuition
Deep Learning has really taken off in the past couple of years . The main reasosns for this exponential growth are majorly -
- Massive increase in computational power and specialised hardware
- Massive amout of useful datasets being generated due to digitalisation of society
But this post is about understanding how algorithmically in their fundamental essence nueral nets work better than traditional ML algorithms.
Many times it may happen that the mapping we are trying to do using traditional algorithms doesnt work very well because real world data problems are not hard and fast defined functions. Thus we need more innovative methods to find really complex decision boundaries.
Nueral Nets are extremely good at finding non linear decision boundaries when given massive data points. Nueral networks can be understood as a complex network of many numbers of nuerons .if each nueron has a activation function as sigmoid function than it is pretty clear that the ouput is not linearly dependent with input and hence the decision boundary is not linear.
And thanks to backpropogation algorithm then the specific weights of each nueron can be adjusted to give highly accurate results.