Maxout Networks: ReLU Is a Maxout
Beginner
The problem, from first principles
Maxout generalizes ReLU: a maxout unit with two pieces, one fixed at zero and one free, reproduces ReLU exactly. The paper uses this to show maxout is at least as expressive as the standard activation.
From theory to code
Implement relu_via_maxout(x), which returns max(x, 0) by taking the maximum of the two pieces x and 0.
Constraints
- Result must equal
np.maximum(x, 0).
Hints
Hint 1
Stack x and an all-zero array along a new last axis, then take the max over that axis.
Keywords
research-papersactivationmaxout
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.