关于用gradient descent 解决graphical model (这里指matrix factorization)的问题
1,根据graphicalmodel,写出distribution之间的关系,找出要minimize的objectivefunction,假设function是f,对应的要解决的matrix有U和V2,用f分别对U和V求解偏导数,得到pian_f/pian_U,pian_f/pian_V,则根据求解出来的公式更新pian_U和pian_V3,对pian_U和pian_V一起进行normalize,