Noisy Gradient Descent Converges to Flat Minima for Nonconvex Matrix Factorization.
Tianyi LiuYan LiSong WeiEnlu ZhouTuo ZhaoPublished in: CoRR (2021)
Keyphrases
- matrix factorization
- missing data
- stochastic gradient descent
- collaborative filtering
- objective function
- low rank
- nonnegative matrix factorization
- cost function
- factor analysis
- negative matrix factorization
- recommender systems
- factorization methods
- low rank matrices
- convex optimization
- loss function
- data matrix
- data sparsity
- probabilistic matrix factorization
- missing values
- implicit feedback
- factorization method
- tensor factorization
- variational bayesian
- matrix completion
- k means
- item recommendation
- clustering algorithm