SGD Noise and Implicit Low-Rank Bias in Deep Neural Networks.
Tomer GalantiTomaso A. PoggioPublished in: CoRR (2022)
Keyphrases
- low rank
- missing data
- neural network
- matrix factorization
- convex optimization
- matrix completion
- low rank matrix
- linear combination
- stochastic gradient descent
- rank minimization
- high order
- kernel matrix
- pattern recognition
- matrix decomposition
- trace norm
- singular value decomposition
- singular values
- missing values
- semi supervised
- high dimensional data
- low rank matrices
- incomplete data
- low rank approximation
- robust principal component analysis
- nearest neighbor
- wavelet transform
- minimization problems
- active learning
- non rigid structure from motion