Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Conformal Prediction: a Unified Review of Theory and New Challenges

Matteo Fontana, Gianluca Zeni, Simone Vantini

TL;DR

This work synthesizes Conformal Prediction (CP), a distribution-free, nonparametric forecasting framework that provides finite-sample valid prediction sets by evaluating how unusual a new instance is relative to past data. It unifies CP theory, practical constructions, and a spectrum of recent developments, including inductive (split) CP, Mondrian validity for category-wise guarantees, and extensions to normalization and functional data bands, all while preserving validity under exchangeability. The survey emphasizes the core idea that a base predictor can be wrapped by a conformal layer to yield guaranteed miscoverage $\alpha$, with efficiency depending on the chosen nonconformity measure and problem setting. The results highlight CP’s broad applicability, computational variants, and adaptability to heteroskedastic and high-dimensional contexts, making it a versatile tool for reliable, distribution-free predictive inference with immediate practical impact.

Abstract

In this work we provide a review of basic ideas and novel developments about Conformal Prediction -- an innovative distribution-free, non-parametric forecasting method, based on minimal assumptions -- that is able to yield in a very straightforward way predictions sets that are valid in a statistical sense also in in the finite sample case. The in-depth discussion provided in the paper covers the theoretical underpinnings of Conformal Prediction, and then proceeds to list the more advanced developments and adaptations of the original idea.

Conformal Prediction: a Unified Review of Theory and New Challenges

TL;DR

This work synthesizes Conformal Prediction (CP), a distribution-free, nonparametric forecasting framework that provides finite-sample valid prediction sets by evaluating how unusual a new instance is relative to past data. It unifies CP theory, practical constructions, and a spectrum of recent developments, including inductive (split) CP, Mondrian validity for category-wise guarantees, and extensions to normalization and functional data bands, all while preserving validity under exchangeability. The survey emphasizes the core idea that a base predictor can be wrapped by a conformal layer to yield guaranteed miscoverage , with efficiency depending on the chosen nonconformity measure and problem setting. The results highlight CP’s broad applicability, computational variants, and adaptability to heteroskedastic and high-dimensional contexts, making it a versatile tool for reliable, distribution-free predictive inference with immediate practical impact.

Abstract

In this work we provide a review of basic ideas and novel developments about Conformal Prediction -- an innovative distribution-free, non-parametric forecasting method, based on minimal assumptions -- that is able to yield in a very straightforward way predictions sets that are valid in a statistical sense also in in the finite sample case. The in-depth discussion provided in the paper covers the theoretical underpinnings of Conformal Prediction, and then proceeds to list the more advanced developments and adaptations of the original idea.

Paper Structure

This paper contains 18 sections, 3 theorems, 34 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Foundations of Conformal Prediction
  3. Conformal Predictors
  4. The Randomness Assumption
  5. Bags and Nonconformity Measures
  6. Conformal Prediction
  7. Validity and Efficiency
  8. Objects and Labels
  9. Classification
  10. Regression
  11. The Online Framework
  12. Recent Advances in Conformal Prediction
  13. Different Notions of Validity
  14. Mondrian Conformal Predictors
  15. Inductive Prediction
  16. ...and 3 more sections

Key Result

Proposition 2.1

Under the exchangeability assumption, the probability of error, $z_{n+1} \notin \gamma^{\alpha}(z_1,\dots,z_{n})$, will not exceed $\alpha$, for any $\alpha$ and any conformal predictor $\gamma$.

Figures (2)

  • Figure 1: Inductive and transductive approach to prediction.
  • Figure 2: Conformal predictors do not contemplate heteroskedasticity in the data distribution. In such a case, one would expect the length of the output interval to be an increasing function of the corresponding variance of the output value, which can give more information of the target label. To tackle this problem, local-weighted conformal inference has been introduced. Source: paper:lei2017.

Theorems & Definitions (3)

  • Proposition 2.1
  • Proposition 2.2
  • Proposition 3.1