AINeutralarXiv – CS AI · 7h ago6/10
🧠
TAROT: Task-Adaptive Refinement of LLM-prior Graphs for Few-shot Tabular Learning
TAROT is a new GNN-based framework that improves few-shot tabular learning by constructing task-adaptive semantic graphs from LLM-inferred feature relationships. The approach addresses privacy concerns of direct LLM tabular data processing while achieving state-of-the-art performance on few-shot benchmarks through intelligent graph refinement that filters LLM hallucinations.