Specific deep learning architectures and architectural components (e.g., layers) are usually designed and/or chosen based on:多项选择题

The target task (e.g., machine translation)

Because that's the way it's always been done

Gut feel

Trying to achieve the best performance possible while also optimizing for efficient training

Trying to minimize overfitting/underfitting

The type of input or output data involved (e.g., sequences, text, images)