Deep InfoMax: The Softplus Function
Beginner
The problem, from first principles
The softplus is a smooth version of ReLU, and it appears throughout mutual information estimators. Written naively, exp overflows for large inputs, so a stable form is needed.
From theory to code
Implement softplus(x), returning log(1 + exp(x)).
Constraints
- Avoid computing
exp(x)directly for large x.
Hints
Hint 1
Use np.logaddexp(0, x), which evaluates the same quantity stably.
Keywords
research-papersunsupervisedmutual-informationrepresentation
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.