Question 5/13 fast.ai v3 lecture 9
The LogSumExp trick above relies on properties of logarithms. We can subtract a number from our
xs and add it at the front without affecting the result. What are the benefits of the LogSumExp?
It helps us address numerical stability by allowing us to operate on much smaller floating point numbers
Relevant part of lecture
A good way of leveraging the LogSumExp trick is to set
a to the max of all