Table 2.

Model network structural hyperparameters

Parameter	Value	Explanation
n_cyc	30	Number of previous cycles used for model input.
batch_size	512	Batch size for training and validation.
Lr	8e-4	Learning rate for training.
num_epochs	12,000	Maximum number of training epochs.
Patience	1,600	Early stopping patience.
Alpha	[0.1] ∗ 10	Capacity loss weight during pre-training.
in_ch	4	Number of input channels for convolution layers.
out_ch	[8, 16, 64]	Number of output channels for convolution layers.
Kernel	3	Kernel size for convolution layers.
Stride	2	Stride for convolution layers.
Padding	0	Padding for convolution layers.
embed_dim	64	Embedding dimension for multi-head attention layers.
num_heads	2	Number of attention heads in multi-head attention layers.
Dropout	0.3	Dropout rate for multi-head attention layers.
dense_1	64	Number of neurons in the first dense layer.
dense_2	64	Number of neurons in the second dense layer.
finetune_lr	2e-5	Learning rate for fine-tuning.
train_alpha	[0.09] ∗ 9 + [0]	Fine-tuning capacity loss weights during training.
valid_alpha	[0.09] ∗9 + [0]	Fine-tuning capacity loss weights during validation.
finetune_epochs	800	Maximum number of fine-tuning epochs.