Today I re-read (for the n-th time) the paper by Bottou, Curtis, Nocedal on the stochastic gradient methods (and similar) seen from an optimization perspective.
Each time I read it I find new insights and I have the feeling of getting closer to understand this fascinating subject.