Table 3 Detailed model architecture and training hyperparameters.

From: physically interpretable residual strength prediction of corroded pipelines via symbolic Bayesian networks

Module

Parameter

Value

BFW-NN

Number of neurons (Hidden Layers)

64, 32, 16

Activation function

ReLU

Dropout rate

0.1

DSR

Number of neurons (Encoder/Decoder)

[8, 64, 32]/[32, 64, 8]

Activation function

Tanh

L1 sparsity coefficient (\(\lambda _s\))

0.005

Training

Batch size

32

Number of epochs

1000

Back to article page