arXiv:2606.25923v1 Announce Type: new Abstract: A digital twin (DT) is a virtual model of a real-world system that can assist decision-making by simulating scenarios induced by different policies. However, typical machine learning-based DTs do not optimise for this use case. We prove that, when model capacity is limited, training DTs to minimise one-step transition errors can produce suboptimal models for ranking sets of policies according to a reward function. We further show that this holds empirically, even with expressive model classes. To address this, we introduce $\text{DT}^2$, a decisi
The proliferation of machine learning-based digital twins highlights the need to optimize their design and training for their primary purpose: assisting decision-making, which is not inherently achieved by merely minimizing simulation errors.
This research addresses a fundamental limitation in digital twin utility, moving them from mere simulators to more effective tools for policy optimization, which is crucial for complex system management.
Digital twins will become more purpose-built for decision support rather than focusing solely on simulation accuracy, potentially changing how they are designed, validated, and integrated into operational systems.
- · Industries relying on complex simulations (e.g., manufacturing, infrastructure)
- · AI developers in simulation and optimization
- · Decision-makers in operational environments
- · Analytics software providers
- · Companies relying on sub-optimal digital twin implementations
- · Academics focused solely on simulation fidelity metrics without decision-making
Improved decision-making efficiency and outcomes in various real-world applications of digital twins.
Accelerated adoption of digital twin technology across diverse sectors due to demonstrably higher value propositions.
The development of a new class of 'decision-native' AI models that prioritize strategic outcomes over pure predictive accuracy in other domains.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG