(a) The data distribution contains critical slices, s1, s2, that represent a small proportion of the dataset. (b) A vanilla neural network correctly learns the general, linear decision boundary but fails to learn the perturbed slice boundary. (c) A user writes slicing functions (SFs), λ1, λ2, to heuristically target critical subsets. (d) The model commits additional capacity to learn slice expert representations. Upon reweighting slice expert representations, the slice-aware model learns to classify the fine-grained slices with higher F1 score.