PowerSGD: Practical Low-Rank Gradient Compression for Distributed Optimization.
Thijs VogelsSai Praneeth KarimireddyMartin JaggiPublished in: NeurIPS (2019)
Keyphrases
- low rank
- convex optimization
- linear combination
- missing data
- matrix factorization
- nuclear norm
- low rank matrix
- semi supervised
- rank minimization
- matrix decomposition
- matrix completion
- singular value decomposition
- kernel matrix
- high dimensional data
- trace norm
- minimization problems
- stochastic gradient descent
- robust principal component analysis
- singular values
- convex relaxation
- image compression
- non rigid structure from motion
- low rank matrices
- neural network
- dimensionality reduction
- denoising
- multiscale