Describe the architecture of U-Net and why skip connections are used there.

U-Net is a convolutional neural network architecture widely used for image segmentation tasks, especially in medical imaging. It consists of two main parts:

1. Contracting (encoder) part — a sequence of convolutional and pooling layers that extract features and reduce spatial resolution.
2. Expanding (decoder) part — a sequence of layers that gradually restore the spatial resolution, using transposed convolutions or upsampling.

Skip connections are direct links between corresponding encoder and decoder layers at the same resolution level. They allow:

• Transferring detailed spatial information from early layers (with high resolution) to the decoder, helping to more accurately delineate object boundaries.
• Improving gradient flow, facilitating training.

Thus, skip connections help the network preserve both global context and local details, which is critical for precise segmentation.