[2605.05218] Horizon-Constrained Rashomon Sets for Chaotic Forecasting

Computer Science > Machine Learning arXiv:2605.05218 (cs) [Submitted on 17 Apr 2026] Title:Horizon-Constrained Rashomon Sets for Chaotic Forecasting Authors:Gauri Kale, Rahul Vishwakarma, Holly Diamond, Ava Hedayatipour, Amin Rezaei View a PDF of the paper titled Horizon-Constrained Rashomon Sets for Chaotic Forecasting, by Gauri Kale and 4 other authors View PDF HTML (experimental) Abstract:Predictive multiplicity and chaotic dynamics represent two fundamental challenges in machine learning that have evolved independently despite their conceptual connections. We bridge this gap by introducing horizon-constrained Rashomon sets, a theoretical framework that characterizes how model multiplicity evolves with prediction horizon in chaotic systems. Unlike static prediction tasks where the Rashomon set remains fixed, chaos induces exponential divergence among initially similar models, fundamentally transforming the nature of predictive equivalence. We prove that the effective Rashomon set contracts exponentially with lead time at a rate determined by the maximum Lyapunov exponent and introduce Lyapunov-weighted metrics that provide tighter bounds on predictive disagreement. Leveraging these insights, we develop decision-aligned selection algorithms that choose among near-optimal models based on downstream utility rather than forecast accuracy alone. Extensive experiments on synthetic chaotic systems (Lorenz-96, Kuramoto-Sivashinsky) and real-world applications (wind power, traff...