Differentiable Learning Under Triage
This work addresses the challenge of optimizing predictive models under triage for applications like content moderation and scientific discovery, but it is incremental as it builds on existing triage concepts with a new algorithm.
The paper tackles the problem of algorithmic triage, where predictive models defer some instances to human experts, by formally characterizing when triage is beneficial and showing that models trained for full automation are suboptimal under triage. It introduces a gradient-based algorithm that finds optimal triage policies and predictive models, outperforming baselines in experiments on content moderation and scientific discovery tasks.
Multiple lines of evidence suggest that predictive models may benefit from algorithmic triage. Under algorithmic triage, a predictive model does not predict all instances but instead defers some of them to human experts. However, the interplay between the prediction accuracy of the model and the human experts under algorithmic triage is not well understood. In this work, we start by formally characterizing under which circumstances a predictive model may benefit from algorithmic triage. In doing so, we also demonstrate that models trained for full automation may be suboptimal under triage. Then, given any model and desired level of triage, we show that the optimal triage policy is a deterministic threshold rule in which triage decisions are derived deterministically by thresholding the difference between the model and human errors on a per-instance level. Building upon these results, we introduce a practical gradient-based algorithm that is guaranteed to find a sequence of triage policies and predictive models of increasing performance. Experiments on a wide variety of supervised learning tasks using synthetic and real data from two important applications -- content moderation and scientific discovery -- illustrate our theoretical results and show that the models and triage policies provided by our gradient-based algorithm outperform those provided by several competitive baselines.