A neural network architecture that uses skip connections to allow information to bypass layers, making it easier to train very deep networks and improving performance.