Conformal Validity Guarantees Exist for Any Data Distribution

As machine learning (ML) gains widespread adoption, practitioners are increasingly seeking means to quantify and control the risk these systems incur. This challenge is especially salient when ML systems have autonomy to collect their own data, such as in black-box optimization and active learning, where their actions induce sequential feedback-loop shifts in the data distribution. Conformal prediction has emerged as a promising approach to uncertainty and risk quantification, but existing variants either fail to accommodate sequences of data-dependent shifts, or do not fully exploit the fact that agent-induced shift is under our control. In this work we prove that conformal prediction can theoretically be extended to \textit{any} joint data distribution, not just exchangeable or quasi-exchangeable ones, although it is exceedingly impractical to compute in the most general case. For practical applications, we outline a procedure for deriving specific conformal algorithms for any data distribution, and we use this procedure to derive tractable algorithms for a series of agent-induced covariate shifts. We evaluate the proposed algorithms empirically on synthetic black-box optimization and active learning tasks.

JAWS: Predictive Inference Under Covariate Shift

We propose \textbf{JAWS}, a series of wrapper methods for distribution-free uncertainty quantification tasks under covariate shift, centered on our core method \textbf{JAW}, the \textbf{JA}ckknife+ \textbf{W}eighted with likelihood-ratio weights. JAWS also includes computationally efficient \textbf{A}pproximations of JAW using higher-order influence functions: \textbf{JAWA}. Theoretically, we show that JAW relaxes the jackknife+'s assumption of data exchangeability to achieve the same finite-sample coverage guarantee even under covariate shift. JAWA further approaches the JAW guarantee in the limit of either the sample size or the influence function order under mild assumptions. Moreover, we propose a general approach to repurposing any distribution-free uncertainty quantification method and its guarantees to the task of risk assessment: a task that generates the estimated probability that the true label lies within a user-specified interval. We then propose \textbf{JAW-R} and \textbf{JAWA-R} as the repurposed versions of proposed methods for \textbf{R}isk assessment. Practically, JAWS outperform the state-of-the-art predictive inference baselines in a variety of biased real world data sets for both interval-generation and risk-assessment auditing tasks.