MAE v.s. MSE as loss function
In the landscape of regression problems, the choice between Mean Squared Error (MSE) and Mean Absolute Error (MAE) as loss functions is pivotal. MSE’s rapid convergence can be an asset, but its susceptibility to outlier influence makes it less suitable for datasets containing anomalies. In contrast, MAE’s robustness against outliers ensures a more balanced training process, ultimately leading to models that generalize better. Understanding the characteristics of your data and the strengths of each loss function is essential in making an informed decision that aligns with your problem’s requirements.
To conclude, now that it is clear the effects outliers can have in squared errors such as MSE or RMSE, it is important to say that in applications which are clear of noise these metrics can do more good than harm, since they can minimize greater errors even though it means accepting more frequent, smaller errors.
intuitively,
1.在回归任务中, 选择 MAE 还是 MSE 作为损失函数? 取决于对数据集纯净程度以及数据集中的 outlier 的分布的情况, 需要思考数据集里面是否有较多的 noise, 以及有多大程度上更关注模型的泛化性
2.在没有 noise 的数据里面, MSE 能够更大限度的减少回归意义上的误差, 使得模型训练过程中具有快速收敛的优势; 但是 MSE 对 outlier 更加敏感
3.MAE 对异常值更加具备鲁棒性, 最终可能使得模型具备更强的泛化能力
Reference
[1]. https://towardsdatascience.com/comparing-robustness-of-mae-mse-and-rmse-6d69da870828
[2]. https://medium.com/@nirajan.acharya666/choosing-between-mean-squared-error-mse-and-mean-absolute-error-mae-in-regression-a-deep-dive-c16b4eeee603#:~:text=In%20the%20realm%20of%20regression,Mean%20Absolute%20Error%20(MAE)
转载请注明来源, from goldandrabbit.github.io
