Mixed-Precision Orthogonalization Scheme and Adaptive Step Size for Improving the Stability and Performance of CA-GMRES on GPUs.
Ichitaro YamazakiStanimire TomovTingxing DongJack J. DongarraPublished in: VECPAR (2014)
Keyphrases
- step size
- variable step size
- adaptive filter
- convergence rate
- cost function
- convergence speed
- cellular automata
- evolutionary programming
- steepest descent method
- hessian matrix
- least mean square
- faster convergence
- gradient method
- quantization step
- image quality
- transform domain
- graphics processing units
- stochastic gradient descent
- line search
- general purpose
- computer vision