Researchers have developed a method to fine-tune language models for crowd simulation by using aggregate mobility statistics, such as zone-level device counts and origin-to-destination (OD) flows, to guide agent behavior. This approach addresses the challenge of under-determined individual behavior when only aggregate data is available. By iteratively fitting the model's destination distribution to observed OD flows, the fine-tuned agents achieved a 25% reduction in destination-share error on data from two baseball games, while maintaining similar grid correlations. AI
IMPACT This research demonstrates a novel application of language models in simulating complex human behavior, potentially improving urban planning and event management.
RANK_REASON The cluster contains a research paper published on arXiv detailing a novel method for applying language models to crowd simulation.
Read on arXiv cs.MA (Multiagent) →
- arXiv
- crowd simulation
- Distilling Aggregate Mobility Statistics into a Language Model Policy for Post-Event Crowd Simulation
- Hugging Face
- language model
- alphaXiv
- CatalyzeX
- CORE Recommender
- DagsHub
- Gotit.pub
- OD flow
- ScienceCast
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
