Discriminative Jackknife: Quantifying Uncertainty in Deep Learning via Higher-Order Influence Functions
This addresses the problem of reliable uncertainty estimation for deep learning practitioners, offering a post-hoc solution that ensures coverage and discriminative accuracy, though it is incremental as it builds on existing jackknife and influence function techniques.
The paper tackles the challenge of quantifying predictive uncertainty in deep learning by developing the discriminative jackknife (DJ), a frequentist method that uses influence functions to construct confidence intervals, achieving competitive performance compared to existing baselines.
Deep learning models achieve high predictive accuracy across a broad spectrum of tasks, but rigorously quantifying their predictive uncertainty remains challenging. Usable estimates of predictive uncertainty should (1) cover the true prediction targets with high probability, and (2) discriminate between high- and low-confidence prediction instances. Existing methods for uncertainty quantification are based predominantly on Bayesian neural networks; these may fall short of (1) and (2) -- i.e., Bayesian credible intervals do not guarantee frequentist coverage, and approximate posterior inference undermines discriminative accuracy. In this paper, we develop the discriminative jackknife (DJ), a frequentist procedure that utilizes influence functions of a model's loss functional to construct a jackknife (or leave-one-out) estimator of predictive confidence intervals. The DJ satisfies (1) and (2), is applicable to a wide range of deep learning models, is easy to implement, and can be applied in a post-hoc fashion without interfering with model training or compromising its accuracy. Experiments demonstrate that DJ performs competitively compared to existing Bayesian and non-Bayesian regression baselines.