Qiwei Li, Jorge Ortiz
Traffic cameras generate mountains of data, but almost all of it is pure observation. Nobody runs experiments (you can't order up rain on Tuesday to see what it does to congestion), which makes causal questions genuinely hard. teLLMe is a tool for poking at those questions carefully. You ask something in plain English, like whether rain changes traffic density, and a schema-aware model turns it into a structured causal query with treatments, outcomes, and subgroups spelled out.
Under the hood it learns a causal graph with the PC algorithm, checks stability with bootstrapping, and estimates effects using regression and the DoWhy library. The output is a "Causal Card" listing the estimate, which variables it adjusted for, and the assumptions behind it, plus a short plain-language note. On traffic events from the BDD dataset it surfaces believable links involving weather and rush hour.
The authors are upfront that this is for generating hypotheses, not settling them. Since I'm going off the abstract, the paper has the case-study specifics.
Traffic agencies now have access to large volumes of video-derived data for studying safety and congestion. Most of these data are observational and collected without interventions, which makes causal questions such as "How would rain change traffic density?" difficult to answer. We present teLLMe, a system for exploratory causal analysis of urban driving datasets. The system starts from a structured event table built from dashcam annotations and combines causal structure…
Frontier Language Models Struggle to Copy: Text Can Be Better Viewed in 2D
arXiv (cs.AI) · July 17, 2026LLM-Powered Agentic AI for 5G/6G Networks: A Tutorial and Survey on Architectures, Protocols, and Standardization
arXiv (cs.AI) · July 17, 2026Spatial Normalization for Cross-Domain Retinal Layer Segmentation in Optical Coherence Tomography
arXiv (cs.AI) · July 17, 2026When Model Merging Rivals Joint Multi-Task Reinforcement Learning: A Task-Vector Geometry Analysis