Distribution-free, Risk-controlling Prediction Sets

2021/01/07 by Stephen Bates, Anastasios N. Angelopoulos, Bates, Stephen +9 · 40 citations
Computer Science · #Machine Learning and Data Classification #Machine Learning and Algorithms #Adversarial Robustness in Machine Learning

paper · pdf · doi:10.1145/3478535

Abstract

While improving prediction accuracy has been the focus of machine learning in recent years, this alone does not suffice for reliable decision-making. Deploying learning systems in consequential settings also requires calibrating and communicating the uncertainty of predictions. To convey instance-wise uncertainty for prediction tasks, we show how to generate set-valued predictions from a black-box predictor that controls the expected loss on future test points at a user-specified level. Our approach provides explicit finite-sample guarantees for any dataset by using a holdout set to calibrate the size of the prediction sets. This framework enables simple, distribution-free, rigorous error control for many tasks, and we demonstrate it in five large-scale machine learning problems: (1) classification problems where some mistakes are more costly than others; (2) multi-label classification, where each observation has multiple associated labels; (3) classification problems where the labels have a hierarchical structure; (4) image segmentation, where we wish to predict a set of pixels containing an object of interest; and (5) protein structure prediction. Last, we discuss extensions to uncertainty quantification for ranking, metric learning, and distributionally robust learning.

Cited by

• Certified Domain Consistency for Multi-Domain Retrieval: Label-Free Per-Domain Contamination Control with Conformal Risk Guarantees
• Combining e-values using demi-supermartingales
• Signed Evidence Flow: Conflict-Aware and Stability-Calibrated Data Analysis
• Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade
• A Step Forward Towards Trustworthy Risk-Aware Facial Retrieval (RA-FR)
• Beyond Aggregate Risk: Role-Stratified Conformal Risk Control for LLM Tool Calls
• A Critical Perspective on Finite Sample Conformal Prediction Theory in Medical Applications
• Uncertainty-Aware Data-Efficient AI: An Information-Theoretic Perspective
• Model-Free Assessment of Simulator Fidelity via Quantile Curves
• A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification
• LEC: Linear Expectation Constraints for Selection-Conditioned Risk Control in Selective Prediction and Routing Systems
• Cost-Sensitive Conformal Training with Provably Controllable Learning Bounds
• Controlling False Positives in Image Segmentation via Conformal Prediction
• Elements of Conformal Prediction for Statisticians
• Consistent Accelerated Inference via Confident Adaptive Transformers
• Conformalized Rate-Adaptive Sensing
• When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents
• Conformal Lesion Segmentation for 3D Medical Images
• Controlling the Risk of Corrupted Contexts for Language Models via Early-Exiting
• On the Provable Performance Guarantee of Efficient Reasoning Models
• Conformal Risk Training: End-to-End Optimization of Conformal Risk Control
• Decomposition-Based Modular Conformal Prediction for Two-Stage Modeling
• Towards Reliable and Holistic Visual In-Context Learning Prompt Selection
• Singleton-Optimized Conformal Prediction
• Beyond Greedy Exits: Improved Early Exit Decisions for Risk Control and Reliability
• Taming Variability: Randomized and Bootstrapped Conformal Risk Control for LLMs
• Conditional predictive inference with L^k-coverage control
• Statistical Inference Leveraging Synthetic Data with Distribution-Free Guarantees
• Statistical Methods in Generative AI
• Synthetic Counterfactual Labels for Efficient Conformal Counterfactual Inference
• Locally Adaptive Conformal Inference for Operator Models
• Reliable Wireless Indoor Localization via Cross-Validated Prediction-Powered Calibration
• QUTCC: Quantile Uncertainty Training and Conformal Calibration for Imaging Inverse Problems
• Few-shot Conformal Prediction with Auxiliary Tasks
• You Don't Bring Me Flowers: Mitigating Unwanted Recommendations Through Conformal Risk Control
• Sensitivity Analysis of Individual Treatment Effects: A Robust Conformal Inference Approach
• Sensitivity analysis of individual treatment effects: A robust conformal inference approach. [europepmc]
• Utilizing risk-controlling prediction calibration to reduce false alarm rates in epileptic seizure prediction. [europepmc]
• Task-Driven Uncertainty Quantification in Inverse Problems via Conformal Prediction. [europepmc]
• Conformal Bounds on Full-Reference Image Quality for Imaging Inverse Problems. [europepmc]

Related

• Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Control
• Semi-Supervised Risk Control via Prediction-Powered Inference
• Data Uncertainty without Prediction Models
• From Data to Uncertainty Sets: a Machine Learning Approach
• Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift