Communication trade-offs for Local-SGD with large step size.
Aymeric DieuleveutKumar Kshitij PatelPublished in: NeurIPS (2019)
Keyphrases
- step size
- stochastic gradient descent
- trade off
- convergence rate
- cost function
- convergence speed
- evolutionary programming
- line search
- adaptive filter
- hessian matrix
- steepest descent method
- variable step size
- gradient method
- faster convergence
- wavelet coefficients
- least squares
- wavelet transform
- multiresolution
- objective function
- image processing