TY  - RPRT
TI  - Lower error bounds for the stochastic gradient descent optimization algorithm: Sharp convergence rates for slowly and fast decaying learning rates
AU  - Arnulf Jentzen
AU  - Philippe von Wurstemberger
PY  - 2018
DO  - 10.1016/j.jco.2019.101438
UR  - https://arxiv.org/abs/1803.08600
ID  - 1803.08600
ER  -