Recreating MATLAB RegressionLayer as custom RegressionLayer?
2 次查看(过去 30 天)
显示 更早的评论
MATLAB Documentation (https://www.mathworks.com/help/deeplearning/ref/regressionlayer.html) says the regression layer using the half-mean-squared-error as the loss function in the regression layer. I'd like to replicate this layer myself, similar to the MAE loss function tutorial (https://www.mathworks.com/help/deeplearning/ug/define-custom-regression-output-layer.html), so that I can ultimately start tweaking it. For now though, I just want to replicate it exactly.
I think the correct (GPU implemented) loss function in the fowardLoss layer would be as follows:
R = size(Y,3); MSE = sum((Y-T).*(Y-T),3)/(R*2); N = size(Y,4); loss = sum(MSE)/N;
However, I'm not sure about the backwardLoss implementation. My best guess would be:
R = size(Y,3); N = size(Y,4); dLdY = (Y-T)/(N*R);
Is this correct?
0 个评论
回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!