A network uses residual connections between layers. What is a primary motivation?

1. Improve optimization by preserving gradient and information paths
2. Replace all learned parameters with fixed labels
3. Remove the need for a loss function
4. Guarantee no overfitting occurs