# Skip Connection
- 
- $x_i = F(x_{i-1}) + x_{i-1}$
- [Effect Of Depth](Effect%20Of%20Depth.md)
- Previous layer gradient carried to next module untouched -> [loss](../Tag%20Pages/loss.md) surface is smoother
- Transfer #architecture to prevent [Vanishingexploding gradients](Vanishingexploding%20gradients.md)
- Learns the difference (residual) $F(x) = H(x)-x$