SCROLL method improves forecasting of stochastic dynamics with learned uncertainty

Researchers have developed SCROLL, a novel method for forecasting multiple observables in stochastic dynamical systems. Unlike traditional approaches that balance per-task losses, SCROLL composes likelihoods of different observables onto a shared backbone, allowing for learned parameterization of unit-dependent loss scaling. This method demonstrates accurate prediction of variance on well-specified systems and separates input-dependent variance on more complex, heteroscedastic systems. SCROLL achieves superior performance in terms of negative log-likelihood and calibration on real-world air-quality data, outperforming tuned baseline methods. AI

IMPACT This research could lead to more accurate and efficient forecasting models for complex systems in fields like climate science and finance.

RANK_REASON The cluster contains an academic paper detailing a new method for forecasting stochastic dynamics. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Lorenz-63
• Ornstein--Uhlenbeck process
• SCROLL

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →