Few-shot Conformal Prediction with Auxiliary Tasks
This work addresses the challenge of reliable uncertainty quantification in few-shot learning scenarios, with applications in NLP, computer vision, and drug discovery.
The paper tackles the problem of conformal prediction with limited training data, which often leads to large prediction sets, by introducing a meta-learning approach over auxiliary tasks, resulting in substantially tighter prediction sets while maintaining marginal guarantees.
We develop a novel approach to conformal prediction when the target task has limited data available for training. Conformal prediction identifies a small set of promising output candidates in place of a single prediction, with guarantees that the set contains the correct answer with high probability. When training data is limited, however, the predicted set can easily become unusably large. In this work, we obtain substantially tighter prediction sets while maintaining desirable marginal guarantees by casting conformal prediction as a meta-learning paradigm over exchangeable collections of auxiliary tasks. Our conformalization algorithm is simple, fast, and agnostic to the choice of underlying model, learning algorithm, or dataset. We demonstrate the effectiveness of this approach across a number of few-shot classification and regression tasks in natural language processing, computer vision, and computational chemistry for drug discovery.