AdaBatch: Efficient Gradient Aggregation.pdf

Westudyanewaggregationoperatorforgradientscomingfromamini-batchforstochastic gradient(SG)methodsthatallowsasignificantspeed-upinthecaseofsparseoptimization problems.WecallthismethodAdaBatchanditonlyrequiresafewlinesofcodechangecom- paredtoregularmini-b