General Description of the Modules section

This section contains basic building blocks that we call “modules”. They are used as bricks to build neural networks.

Variables and Attributes

Each module can perform pure calculations based only on a small number of parameters specified during its creation, as well as contain variables or attributes.

The difference:

  • You can always compute a gradient on variables.
  • Attributes are everything that is significant and requires support for save-load. The base interface of all modules is defined by the Module class.

List of modules