Conformal prediction for the design problem

Abstract

In many real-world deployments of machine learning, we use a prediction algorithm to choose what data to test next. For example, in the protein design problem, we have a regression model that predicts some real-valued property of a protein sequence, which we use to propose new sequences believed to exhibit higher property values than observed in the training data. Since validating designed sequences in the wet lab is typically costly, it is important to know how much we can trust the model's predictions. In such settings, however, there is a distinct type of distribution shift between the training and test data: one where the training and test data are statistically dependent, as the latter is chosen based on the former. Consequently, the model's error on the test data-that is, the designed sequences-has some non-trivial relationship with its error on the training data. Herein, we introduce a method to quantify predictive uncertainty in such settings. We do so by constructing confidence sets for predictions that account for the dependence between the training and test data. The confidence sets we construct have finite-sample guarantees that hold for any prediction algorithm, even when a trained model chooses the test-time input distribution. As a motivating use case, we demonstrate how our method quantifies uncertainty for the predicted fitness of designed protein using several real data sets.

Conformal Risk Control

“…We primarily build on split conformal prediction [2]; statistical properties of this algorithm including the coverage upper bound were studied in [8]. Recently there have been many extensions of the conformal algorithm, mainly targeting deviations from exchangeability [9][10][11][12] and improved conditional coverage [3,[13][14][15][16]. Most relevant to us is recent work on risk control in high probability [17][18][19] and its applications [20][21][22][23][24][25][26].…”

Recommendation Systems with Distribution-Free Reliability Guarantees

“…Our approach is based on recent developments in conformal prediction [Vovk et al, 1999, Papadopoulos et al, 2002, Vovk et al, 2005 and distribution-free uncertainty quantification more broadly [Park et al, 2020, Bates et al, 2021a. This line of work provides a formal approach to defining set-valued statistical predictions and it has been applied to various learning tasks, such as distribution estimation [Vovk et al, 2020], causal inference Candès, 2020, Jin et al, 2021], weakly-supervised data [Cauchois et al, 2022], survival analysis , design [Fannjiang et al, 2022], model cascades [Fisch et al, 2020, the few-shot setting , handling dependent data [Chernozhukov et al, 2018, Dunn et al, 2020, and handling or testing distribution shift [Tibshirani et al, 2019, Cauchois et al, 2020, Hu and Lei, 2020, Bates et al, 2021b, Gibbs and Candès, 2021, Vovk, 2021, Podkopaev and Ramdas, 2022. Most closely related to the present work, there have been recent proposals applying conformal prediction to recommender systems.…”

Improving Trustworthiness of AI Disease Severity Rating in Medical Imaging with Ordinal Conformal Prediction Sets

“…In particular, conformal prediction methods have been applied to a wide range of safety-critical applications -from reducing false alarms in the detection of sepsis risk [29] to end-to-end autonomous driving control [22]. Distribution-free uncertainty quantification techniques such as conformal prediction sets have emerged as an essential tool for rigorous statistical guarantees in medical decision-making [9,20,4,30,11,19].…”