The Step Decay Schedule: A Near Optimal, Geometrically Decaying Learning Rate Procedure For Least Squares.
Rong GeSham M. KakadeRahul KidambiPraneeth NetrapalliPublished in: NeurIPS (2019)
Keyphrases
- learning rate
- least squares
- learning algorithm
- error function
- convergence rate
- hidden layer
- rapid convergence
- scheduling problem
- convergence speed
- adaptive learning rate
- multilayer neural networks
- robust estimation
- training algorithm
- weight vector
- machine learning
- levenberg marquardt
- optical flow
- fixed point
- genetic programming
- delta bar delta