Roughly speaking, the input to the neural network is a list of football teams, the middle layers should contain statistics (how many passes, corners), and the output is the result (how many goals). Is it possible to do such intermediate training, and so that the last layers do not ignore the logits from the first layer? And they are also taken into account at the last layer.

Zenit        [101010]    Passes       [101010]
Spartak   => [001101] => Corners   => [101010] => Number of goals
Dynamo       [111010]    Penalties    [101010]
                         Free kicks   [101010]

I tried googling but couldn’t find anything