AdaGrad optimizer provably works with messy, extreme gradient noise

Original: Can Adaptive Gradient Methods Converge under Heavy-Tailed Noise? A Case Study of AdaGrad

Source: arxiv.org ↗