Scaling Multinomial Logistic Regression via Hybrid Parallelism
pp. 1460-1470, 2019.
We present a novel distributed stochastic optimization algorithm DS-MLR to solve multinomial logistic regression problems having large number of examples and classes
We study the problem of scaling Multinomial Logistic Regression (MLR) to datasets with very large number of data points in the presence of large number of classes. At a scale where neither data nor the parameters are able to fit on a single machine, we argue that simultaneous data and model parallelism (Hybrid Parallelism) is inevitable. ...More
PPT (Upload PPT)