Reproduce “Continuously Updating Digital Twins using Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Harry Amad, Nicolás Astorga, Mihaela van der Schaar
CALM-DT, a Context-Adaptive Language Model-based Digital Twin, enables continuous updates and adaptation to dynamic modeling environments without requiring architectural redesign or parameter retraining. This framework achieves superior simulation accuracy in medical and ecological scenarios and seamlessly incorporates novel actions and continuously updated data/knowledge.
Key insights
- LLMs can function as flexible digital twins, accommodating varying state-action dimensions and enabling continuous learning without requiring explicit parameter updates or re-training.
- Effectively managing the LLM's context window through dynamic selection of relevant in-context examples and domain knowledge is crucial for accurate and long-horizon simulations in complex systems.
- Fine-tuning retrieval encoders with LLM simulation performance feedback and enriching trajectory representations with LLM-generated summaries significantly improves the utility and relevance of retrieved context samples.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.