mlsanityData to 7 Oct 2026

Conformal Policy Control

arXiv 2603.02196 ↗First posted 2 Mar 2026cs.AI cs.LG math.STcited by 5

Not on the map. Why?

Abstract (arXiv)

An agent must try new behaviors to explore and improve. In high-stakes environments, an agent that violates safety constraints may cause harm and must be taken offline, curtailing any future interaction. Imitating old behavior is safe, but excessive conservatism discourages exploration. How much behavior change is too much? We show how to use any safe reference policy as a probabilistic regulator for any optimized but untested policy. Conformal calibration on data from the safe policy determines how aggressively the new policy can act, while provably enforcing the user's declared risk tolerance. Unlike conservative optimization methods, we do not assume the user has identified the correct model class nor tuned any hyperparameters. Unlike previous conformal methods, our theory provides finite-sample guarantees even for non-monotonic bounded loss functions, and it introduces a new policy control setting. Our experiments on applications ranging from natural language question answering to biomolecular engineering show that safe exploration is not only possible from the first moment of deployment, but can also improve performance.

Key facts. Cited by 5 papers in the map (data through 7 Oct 2026). Builds on 80 of its 125 references that are in the map. First posted on arXiv 2 Mar 2026.

Builds on

80 of 125 references.

References · 80

  1. 2017Proximal Policy Optimization Algorithmscited by 14,441
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  3. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,115
  4. 2017Deep reinforcement learning from human preferencescited by 3,846
  5. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  6. 2015Trust Region Policy Optimizationcited by 2,503
  7. 2020Conservative Q-Learning for Offline Reinforcement Learningcited by 1,777
  8. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,734
  9. 2016Concrete Problems in AI Safetycited by 1,582
  10. 2019Conformal Prediction Under Covariate Shiftcited by 511
  11. 2019The frontier of simulation-based inferencecited by 490
  12. 2021Adaptive Conformal Inference Under Distribution Shiftcited by 360
38 more
  1. 2019Scalable Global Optimization via Local Bayesian Optimizationcited by 359
  2. 2022Conformal prediction beyond exchangeabilitycited by 348
  3. 2022Conformal Risk Controlcited by 290
  4. 2021Distribution-Free, Risk-Controlling Prediction Setscited by 278
  5. 2021Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Controlcited by 256
  6. 2024Adjoint Matching: Fine-tuning Flow and Diffusion Generative Models with Memoryless Stochastic Optimal Controlcited by 222
  7. 2019Predictive inference with the jackknife+cited by 206
  8. 2023Conformal Language Modelingcited by 163
  9. 2022Adaptive Conformal Predictions for Time Seriescited by 155
  10. 2020Conformal Inference of Counterfactuals and Individual Treatment Effectscited by 150
  11. 2024Fine-Tuning of Continuous-Time Diffusion Models as Entropy-Regularized Controlcited by 136
  12. 2016Bayesian Optimization with Safety Constraints: Safe and Automatic Parameter Tuning in Roboticscited by 134
  13. 2024Language Models with Conformal Factuality Guaranteescited by 128
  14. 2015Taming the Noise in Reinforcement Learning via Soft Updatescited by 125
  15. 2022Safe Planning in Dynamic Environments using Conformal Predictioncited by 112
  16. 2020Telescoping Density-Ratio Estimationcited by 108
  17. 2021Learning Optimal Conformal Classifierscited by 108
  18. 2024Large language model validity via enhanced conformal prediction methodscited by 91
  19. 2021Conservative Objective Models for Effective Offline Model-Based Optimizationcited by 87
  20. 2015Safe Controller Optimization for Quadrotors with Gaussian Processescited by 77
  21. 2019Adaptive and Safe Bayesian Optimization in High Dimensions via One-Dimensional Subspacescited by 73
  22. 2022Selection by Prediction with Conformal p-valuescited by 66
  23. 2022Practical Adversarial Multivalid Conformal Predictioncited by 65
  24. 2022Improving Expert Predictions with Conformal Predictioncited by 54
  25. 2022Conformal Prediction Under Feedback Covariate Shift for Biomolecular Designcited by 53
  26. 2023How to Trust Your Diffusion Model: A Convex Optimization Approach to Conformal Risk Controlcited by 50
  27. 2023Conformal Decision Theory: Safe Autonomous Decisions from Imperfect Predictionscited by 46
  28. 2021Sample-Efficient Safety Assurances using Conformal Predictioncited by 41
  29. 2022Doubly Robust Calibration of Prediction Sets under Covariate Shiftcited by 40
  30. 2022Achieving Risk Control in Online Learning Settingscited by 38
  31. 2025Decision Theoretic Foundations for Conformal Prediction: Optimal Uncertainty Quantification for Risk-Averse Agentscited by 37
  32. 2022Bayesian Optimization with Conformal Prediction Setscited by 27
  33. 2020Towards Safe Policy Improvement for Non-Stationary MDPscited by 26
  34. 2023Model-free selective inference under covariate shift via weighted conformal p-valuescited by 26
  35. 2024Conformal Validity Guarantees Exist for Any Data Distribution (and How to Find Them)cited by 26
  36. 2024Formal Verification and Control with Conformal Predictioncited by 26
  37. 2022Conformal Off-Policy Prediction in Contextual Banditscited by 25
  38. 2020Training conformal predictorscited by 20

50 most cited of 80 All 80 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026Online Reasoning Calibration: Test-Time Training Enables Generalizable Conformal LLM Reasoningcited by 4
  2. 2026Audited Selective Verification for Risk-Controlled N-1 Thermal Contingency Screening under Deployment Shiftcited by 1
  3. 2026Selective Safety Steering via Value-Filtered Decodingcited by 0
  4. 2026Reference-Tail Trust:Certified Probability Floors for Learned Updates Inside a Deployed Networkcited by 0
  5. 2026Safe Meta-Policy Design with Risk Controlcited by 0