Parametric ReLU: Learning the Negative Slope
Beginner
The problem, from first principles
ReLU sets every negative input to zero, which kills gradients for those units. Parametric ReLU (He et al., 2015) keeps a small learnable slope a for negative inputs, so those units still carry signal and gradient.
From theory to code
Implement prelu(x, a), which returns x for positive entries and a * x for non-positive ones.
Constraints
ais a float;a = 0gives ordinary ReLU.
Hints
Hint 1
np.where(x > 0, x, a * x) applies the rule element-wise in one call.
Keywords
research-papersactivationrelu
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.