Wang, Y. Risk-Aware Large Language Model Fine-Tuning With Tail-Robust Objectives and Feasible Gradient Updates. JCTS 2025, 4.