Conformal Prediction Regions for Continuous-Time Trajectories under Random Sampling
arXiv:2608.29559
2026
Sampling
2 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper provides a constructive way to extend split-conformal coverage from randomly sampled trajectory points to an entire continuous-time path by combining a discrete conformal residual with a bound on trajectory regularity. Its key transferable asset is a deterministic gap-inflation rule: uncertainty between observations grows proportionally to the largest time gap and a Lipschitz or derivative bound. This can wrap neural ODEs, state-space models, and world models to produce safety-relevant continuous-time prediction tubes despite sparse or random observations. A second transfer is an adaptive sampling controller that refines solver or sensor time points whenever the conformal tube becomes too wide.
Ideas from this paper
△ Mechanism confirmed, baseline not beaten
2026
Wrap a neural ODE, recurrent state-space model, or learned world model with a split-conformal prediction tube that is valid between irregularly sampled observations. Calibrate a pointwise residual quantile at observed times and inflate it at an unobserved time according to its distance from the nearest observed time and an estimated bound on the true and predicted trajectory slopes.
Useful8/10
Difficulty4/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Use the conformal regularity inflation law as a controller for observation placement or neural-ODE solver refinement. Sample or evaluate the learned dynamics more densely only where the predicted continuous-time uncertainty exceeds a prescribed safety radius, rather than using a uniform time grid.
Useful7/10
Difficulty5/10
Novelty8/10