在matlab中如何更改cnn神经网络训练损失函数

    properties(SetAccess = private)
        % LossFunction   The loss function for training
        %   The loss function that will be used during training. Possible
        %   values are:
        %       'crossentropyex'    - Cross-entropy for exclusive outputs.
        LossFunction = 'focalloss';//默认值为'crossentropyex'
    end

在类定义ClassificationOutputLayer中将LossFunction更改为你都函数名即可，这里我使用focalloss作为我都损失函数，以应对在卷积神经网络分类中，样本不平衡问题。prediction和target分别为预测值和真实值。

function loss = focalLoss(prediction, target)
    gamma = 2;
    alpha = 0.25;
    ce = -target .* log(prediction);
    pt = prediction;
    alpha_term = alpha * target .* ((1 - pt) .^ gamma);
    loss = mean(ce .* alpha_term);
end