Research Article
Online Learning for DNN Training: A Stochastic Block Adaptive Gradient Algorithm
| | Input: | | | Parameter: , and where and . denotes coordinate selection probability at time . Moreover, where and . | | | Initially Set: and . | | | Output: | | (1) | fordo | | (2) | | | (3) | Generating diagonal matrix with probability | | (4) | | | (5) | Generating gradient | | (6) | | | (7) | | | (8) | and | | (9) | Clip | | (10) | | | (11) | | | (12) | end for | | (13) | return |
|