AutoRec은 RBM-based CF model과 차이가 있음
- 기저
- RBM-based model은 generative, probabilitstic model based on boltzmann machines
- AutoRec은 discriminative model based on autoencoders
- 최적화
- RBM-CF estimates parameters by maximaize log likelihood
- AutoRec directly minimize RMSE
- 학습
- RBM-CF require contrastive divergence
- AutoRec comparatively faster gradient-based backpropagation
- 파라미터 개수
- RBM-CF nkr or mkr
- AutoRec need fewer parameter
- Compare to MF, MF need user, item hidden vector
- AutoRec only need Item latent vector
- 선형
- MF learns a linear latent representation
- AutoRec learn a nonlinear latent representation