A Unified Probabilistic Framework for Prediction Under Uncertainty in AI Drug Discovery and Hypersonic Object Tracking

Read full text Summarise with VIRAT
Authors: Dimitar Kretski
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

AbstractWe propose a unified framework for AI-based controller discovery with built-in physics and Lyapunov-based guarantees. We introduce a functional solution (W(t)=Q(t)\cdot D(t)T(t)) and prove that under the explicitly stated assumptions there exist constants (k>0,\beta\ge0) such that (\dot V\le -kW(t)+\beta), which provide formal conditions for robustness and convergence in a simulation model. We present the ORAC (Physics Constrained Evolutionary Controller Discovery) algorithm, which combines candidate generation, physics consistency filters, and Lyapunov scoring for early rejection of uncertain controllers. Demonstrate the applicability of the approach through three keys: ORAC NT (theoretical coverage and simulation verification), Tokamak/plasma (simulation success, hardware validation required), and Hypersonic Tracking (empirical improvement without formal guarantees). The results show that Lyapunov induction is an effective inductive bias for safe and reproducible controller discovery, provided that the assumptions are met._______________________________________IntroductionMotivationAutonomous controller discovery through AI accelerates the design of complex dynamical systems, but the lack of formal guarantees limits its adoption in critical applications. All this work is about integrating classical Lyapunov theory with modern generative and evolutionary methods, so that controllers are discovered that can be mathematically verified within the model.Main components• Solution functional (W(t)) as a criterion for selecting and ranking candidates.• Lyapunov score for formal verification of stability in simulation.• ORAC pipeline: generation → simulation → physics verification → Lyapunov filter → evolution.Contributions1. We formulate and analyze a solution functional (W(t)) and its relationship with Lyapunov energy reduction.2. We prove conditions for stability and convergence in the simulation model through the inequality (\dot V\le -kW(t)+\beta).3. We propose a practical algorithm for constrained physical detection of controllers and demonstrate its applicability in three different domains, clearly distinguishing formal guarantees from empirical observations.

Fields of Study & Subjects

Evolutionary Algorithms and Applications Advanced Control Systems Design Advanced Control Systems Optimization

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN
Document type
Preprint