Learning to Defer with an Uncertain Rejector via Conformal Prediction

We perform uncertainty quantification for the rejector sub-component of the learning-to-defer framework. In particular, we use conformal prediction to allow the reject to output sets, instead of just the binary outcome of 'defer' or not. Our method builds on the existing L2D framework, which previously optimized a specific surrogate loss, namely the one-over-all loss. We evaluate our approach by reporting abstention and consensus prediction results on the CIFAR-10 dataset, demonstrating improvements over the traditional method.