Research Article
Robust and Adaptive Online Time Series Prediction with Long Short-Term Memory
| RoAdam. Parameters carried over from Adam have the same default values: , | | , , . For parameters specific to our method, we recommend | | default values , , . | | Require: : learning rate | | Require: : exponential decay rates for moment estimation in Adam | | Require: : exponential decay rate for computing relative prediction error | | Require: , : lower and upper threshold for relative prediction error | | Require: : fuzz factor | | Require: : loss function | | Require: : initial value for parameters | | | | | t = 0 | | while stopping condition is not reached do | | | | | | | if then | | | else | | | end if | | | | | end while | | return |
|
