I am training an ANN model which must produce positive predictions ( e.g. length of some objects ). The way I am planning to implement this constraint is to attach a one-node layer (output layer) at the end to an existing neural network and designate zero bias to this output node. I wonder how I can 'fix' the weights and biases for nodes/layers during training process.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…