learning to reweight examples (lre)

A technique to adaptively reweight the loss function in each mini-batch, useful in scenarios where training examples are noisy or have varying quality.