Dropout: The Forward Pass With Inverted Scaling
Beginner
The problem, from first principles
Dropout (Srivastava et al., 2014, following Hinton et al., 2012) fights overfitting by randomly switching off units during training. Each unit is kept with probability 1 - p and dropped with probability p. Without correction, the next layer would see activations that are too small on average, so the survivors are scaled up. This question implements that forward pass.
From theory to code
Implement dropout_forward(x, keep_mask, p). It multiplies x by the mask and rescales by 1 / (1 - p). The signature and docstring are already in the editor.
Constraints
keep_maskhas the same shape asxand holds 0s and 1s.0 <= p < 1.
Hints
Hint 1
Multiply element-wise by the mask first, then divide by 1 - p.
Keywords
research-papersregularizationdropout
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.