Why do we come up with idea of neural networks and use it?
Why can’t we just run one logistic regression(which I assume is intuitively faster) that running 3 layers with 3 units each which perform logistic regression individually…
kindly help me understand this one…
With regards,
Maran
The key benefit of an neural network is that it has a hidden layer, which uses a non-linear activation function. This allows the NN to automatically learn complex non-linear relationships between the input features.
Right! It is true that Logistic Regression is less expensive to train and to execute, but it has a really fundamental limitation: it can only express linear decision boundaries. If you think about the geometric interpretation of Logistic Regression, the combination of w and b define a hyperplane in the input space. All “Yes” answers are on one side of the hyperplane and all “No” answers are on the other side. w is the normal vector to the plane and b expresses the perpendicular distance from the origin.
It works well if your actual data has a decision boundary that is at least “close” to linear, but for a lot of datasets that just is too big a limitation.
As Tom points out, an NN gives you the ability to “compose” a series of non-linear functions (one for each layer) and that gives you the ability to express a much more complex non-linear decision boundary.
Thank you Mr.Paul
Understood…