ML2.2

Going further with Machine Learnia 2, we see how an activation function

enables us to assign probabilities to outputs.

With the z value on the x-axis, the sigmoid function tells us how probably toxic or not

a given plant smaple is. Below, 10 % probability of being toxic, which is equivalent

to 90% probability of being non toxic!

Substituting:

                                                                           

Thus far, we have seen the following relationships:

                                                    

What we are missing are the two w values, or weights and the b value!

But here's the thing: probability and likelihood are not the same thing. Conditional probabilities

add up to 1; likelihood need not. The machine needs to get as much information as possible

from the data is has at any one point...

                                                               *     *     *

About Linear Regression | IBM

What is Gradient Descent? | IBM