Does setting BiasLearnRateFactor=0 prevent bias updates in a fullyConnectedLayer?

1 次查看（过去 30 天）
태석
태석 2025-7-11
回答： Matt J 2025-7-11
actorNet = [
featureInputLayer(prod(obsInfo.Dimension))
fullyConnectedLayer(64, Bias=zeros(64,1), BiasLearnRateFactor=0)
reluLayer
fullyConnectedLayer(64, Bias=zeros(64,1), BiasLearnRateFactor=0)
fullyConnectedLayer(prod(actInfo.Dimension))
scalingLayer(Scale=5000, Bias=0)
];
actorOptions = rlOptimizerOptions( ...
LearnRate=1e-4, ...
GradientThreshold=1, ...
L2RegularizationFactor=1e-4);
I constructed an actor network and set the optimizer options as shown above.
My question is:
If I set BiasLearnRateFactor=0 in the fullyConnectedLayer, does this mean that the bias will not be updated at all during training?

请先登录，再回答此问题。

回答（1 个）

Matt J
Matt J 2025-7-11
Yes.

类别

在 Help Center 和 File Exchange 中查找有关 Matrix Computations 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by