Improving machine learning with an old approach

Rong Ge is an assistant professor of computer science. Computer scientist Rong Ge has an interesting approach to machine learning. While most machine learning specialists will build an algorithm which molds to a specific dataset, Ge builds an algorithm which he can guarantee will perform well across many datasets. A paper he wrote as a postdoc at Microsoft Research, Escaping From Saddle Points—Online Stochastic Gradient for Tensor Decomposition, describes how a programmer can use the imprecision of a common machine learning algorithm, known as stochastic gradient descent, to his advantage. Normally this algorithm is used to speed up a slow learning process by only approximating the correct solution rather than working harder to get precision; however, Ge and his colleagues found that the small amount of “noise” created by the…


Link to Full Article: Improving machine learning with an old approach