Using downhill simplex method for optimizing machine learning training running time

Many modern machine learning algorithms rely on a set of configuration parameters, such that a proper setting of their values influences the accuracy and the running time of the algorithm. We propose an automated approach based on the Downhill Simplex optimization method for calculating the optimal parameter set in terms of training time. We demonstrated 5X-30X speedups for training the text analytics algorithm word2vec, just through using better configuration parameters.