It is trained once by actual error in each frame to reduce the estimateerror.
CMAC每帧用实际误差训练一次，以减少其预测误差。
2
Although large workload, the final conclusion usage is uncertain for gross estimateerror can't be calculated caused by investigation not being strictly random.
The filter factors calculated based on the available quantile statistics are not exact, but the estimateerror is below 5%, even when only 20 values are collected.