Oriel — AI & technology worth noticing, every day.

Calibrated for Which Policy? Conformal Uncertainty for Fixed World Models Under Intervention

Paper · https://arxiv.org/abs/2610.35877

arXiv:2610.35877v1 Announce Type: new Abstract: Changing a deployment policy can invalidate a world model's uncertainty while leaving the physical transition mechanism and predictor unchanged. For one-step prediction, if source and target laws share the conditional mechanism $P(Y|X)=Q(Y|X)$, then the intervention is identified by the pre-outcome change of measure $w(X)=dQ_X/dP_X$. For policy-controlled dynamics this ratio factors into state-occupancy and action-propensity terms. Weighted conformal calibration transports marginal coverage to the target law; with an estimated ratio $\hat w$, $P_Q\{Y \in C_{\hat w}(X)\} \ge 1-\alpha-TV(Q,\hat Q) \ge 1-\alpha-E_P|w-\hat w|$. Population log-loss regret for a source-versus-target domain classifier bounds the same ratio error. If $h$ is its clipped target posterior, $\rho$ is the target-domain prior, and $\Delta(h)$ is population log-loss regret, then $E_P|w-\hat w_h| \le (1/(\rho\gamma))\sqrt{\Delta(h)/2}$. An exact cancellation improves the dependence from a generic odds-map bound of order $\gamma^{-2}$ to $(\rho\gamma)^{-1}$. Our experiments use a released DINO-feature world model and physical Franka trajectories. Under the strongest controlled state-action shift, ordinary 90% conformal sets cover 55.0%, whereas oracle and learned-ratio calibration cover 91.0% and 90.9%. The learned surrogate remains within 0.9% total variation of the target law. Calibration is relative to the deployment law: the model can remain fixed while calibration follows the intervention. We machine-check in Lean 4 the finite-support algebraic and causal identities used in the classifier-regret argument, including its coverage-composition step.

First seen by Oriel on .

AI summary

This paper shows weighted conformal calibration can preserve uncertainty coverage when deployment policies change, using estimated state-action density ratios while keeping the world model fixed.

  • Under shared conditional dynamics, intervention effects reduce to a pre-outcome density ratio that factors into state occupancy and action propensity.
  • Weighted conformal calibration guarantees target coverage at least 1−α−TV(Q,hat Q), with ratio error bounded by E_P|w−hat w|.
  • A target-domain classifier’s population log-loss regret bounds ratio error by (ργ)^−1√(Δ(h)/2), improving a generic γ^−2 dependence.
  • In controlled Franka trajectories with a released DINO-feature world model, ordinary 90% sets covered 55.0%, versus 91.0% oracle and 90.9% learned-ratio calibration.
  • The learned surrogate stayed within 0.9% total variation of the target law, and Lean 4 checked the finite-support algebraic and causal identities.

#conformal prediction #world models #distribution shift #lean 4

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
arXiv AI Papersoai:arXiv.org:2610.35877v1https://arxiv.org/abs/2610.35877 Read discussion