Bayesian parameter estimation with guarantees via interval analysis and simulat...
summary
Method computes provable Bayesian posterior estimates for models y = f(θ) + h(ψ) (covers noisy ODE parameters and neural-network f) by combining P-boxes, Interval Arithmetic and Monte Carlo into an efficient randomized algorithm that returns confidence intervals for posterior CDF values and moments when the observation lies in a small rectangle. Implementation compares against state-of-the-art on ODE benchmarks and shows marked gains in accuracy and runtime while providing finite-sample guarantees of correctness.
AI-written summary · about