Mixup Loss on Mixed Probabilities

Problem

Compute mixup loss for a pair of samples: mix the probability vectors as lam*probs_a + (1-lam)*probs_b, then return lam*CE(y_a) + (1-lam)*CE(y_b) on the mixed distribution, clipping probabilities to at least 1e-15.

lam is the mixing weight in [0, 1].

Hint

Mix both the inputs (proxied by the probabilities) and the targets with the same lambda.

Test cases

case 1

in: [0,1,[0.8,0.2],[0.1,0.9],0.7]

expected: 0.6368223552427954

case 2

in: [0,1,[0.8,0.2],[0.1,0.9],1]

expected: 0.2231435513142097

case 3

in: [1,0,[0.3,0.7],[0.6,0.4],0.5]

expected: 0.6981723484866961

Starter codePython

import math
def mixup_loss(y_a, y_b, probs_a, probs_b, lam):
    # Your code here
    pass

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