Deep learning
Residual connection
A residual connection adds a block’s input to its transformed output.
How it works
The identity path lets information and gradients bypass a difficult transformation. This makes very deep networks easier to optimize.
A concrete example
Illustrative: input [1.0, 2.0] plus block output [0.2, -0.3] gives [1.2, 1.7].
Do not confuse it with Dense connection
residual paths add tensors; dense connections concatenate features.