上QQ阅读APP看书，第一时间看更新
L1 loss
The L1 loss function, also known as the mean absolute error, measures the average point-wise difference between the model prediction, 
, and the target value, 
. The partial derivative is equal to 1 when the model prediction is larger than the target value, and equal to -1 when the prediction is smaller than the target error. This property of the L1 loss function can be used to circumvent problems that might arise when learning from noisy labels:
