Dead Neuron Fraction

Problem

Compute the fraction of dead neurons in a layer. activations is a (num_neurons, num_examples) matrix; a neuron is dead when none of its activations is strictly positive.

The signature is dead_neuron_fraction(activations).

Hint

Dead ReLU units never recover once their pre-activations stay negative.

Test cases

case 1

in: [[[0,0,0],[1,0,0],[-1,1,0]]]

expected: 0.3333333333

case 2

in: [[[0]]]

expected: 1

case 3

in: [[[1],[2]]]

expected: 0

case 4

in: [[[0,0],[0,1],[0,0],[0,-1]]]

expected: 0.75

Starter codePython

def dead_neuron_fraction(activations):
    # Your code here
    pass

4 test cases · run your code against them in the editor