Can Adaptive Gradient Methods Converge under Heavy-Tailed Noise? A Case Study of AdaGrad math.OC · 2026-05-18