Broadcasting

Some nodes, such as Add or Sub, supports broadcasting the inputs to a common shape. This means that if the two inputs do not have the exact same shape, the shapes will be modified so that the two shapes match and the operation can be performed. You can also use the node Broadcast if you want to broadcast a tensor to a specific shape.

Details of how this works can be found here: Numpy: Broadcasting, and here: PyTorch: Broadcasting semantics.