Deep Linear Networks Dynamics: Low-Rank Biases Induced by Initialization Scale and L2 Regularization.
Arthur JacotFrançois GedFranck GabrielBerfin SimsekClément HonglerPublished in: CoRR (2021)
Keyphrases
- low rank
- rank minimization
- trace norm
- low rank matrix
- minimization problems
- kernel matrices
- matrix factorization
- convex optimization
- linear combination
- missing data
- group sparsity
- matrix completion
- singular value decomposition
- kernel matrix
- matrix decomposition
- rank constraint
- high order
- robust principal component analysis
- semi supervised
- high dimensional data
- low rank approximation
- singular values
- biological networks
- low rank matrices
- pairwise
- complex networks
- markov random field
- k means