10 papers
Execution-First Synthetic Tool-Use Trace Generation for LLM Agents
Hafsa Ouajdi, Francesco Giannuzzo, Alaa Boukhary +3
Agentic software-engineering and industrial systems increasingly operate through executable workflows rather than code genera- tion alone: they search artifacts, invoke tools, insp…
Latent Bridges for Multi-Table Question Answering
Simone Varriale, Tamara Cucumides, Floris Geerts +1
We introduce GRAB, a constructor-encoder-bridge pipeline for table question answering. Our method lifts relational data into an heterogeneous graph, encodes it via message passing,…
Selectivity Estimation for Semantic Filters on Image Data
Matthias Urban, Vu Huy Nguyen, Gabriele Sanmartino +2
Semantic data systems integrate Large Language Models (LLMs) and Vision-Language Models (VLMs) directly into database query execution, enabling expressive queries on multi-modal da…
Constraint Decay: The Fragility of LLM Agents in Backend Code Generation
Francesco Dente, Dario Satriani, Paolo Papotti
Large Language Model (LLM) agents demonstrate strong performance in autonomous code generation under loose specifications. However, production-grade software requires strict adhere…
The Stretto Execution Engine for LLM-Augmented Data Systems
Gabriele Sanmartino, Matthias Urban, Paolo Papotti +1
LLM-augmented data systems enable semantic querying over structured and unstructured data, but executing queries with LLM-powered operators introduces a fundamental runtime-accurac…
Variable Selection in Maximum Mean Discrepancy for Interpretable Distribution Comparison
Kensuke Mitsuzawa, Motonobu Kanagawa, Stefano Bortoli +2
We study two-sample variable selection: identifying variables that discriminate between the distributions of two sets of data vectors. Such variables help scientists understand the…