NCA-GENM Question 21
Single answerA team is designing a generative AI model for image synthesis using a neural network architecture. They decide to incorporate residual connections into the model. What is the primary benefit of using residual connections in nonsequential neural networks?
- A
They prevent overfitting by adding dropout layers to the network.
- B
They allow the network to learn identity mappings, improving gradient flow during training.
- C
They reduce the computational cost of training by skipping layers entirely.
- D
They ensure that each layer processes inputs sequentially for better feature extraction.
Show answer and explanation
Correct answer: B
Explanation
Residual connections are critical for deep neural networks, especially nonsequential architectures, as they enable the network to learn identity mappings. This improves gradient flow during training and helps mitigate challenges like vanishing gradients, making it easier to train very deep networks.
- A. Incorrect.
Residual connections are not directly related to preventing overfitting; dropout layers are used for this purpose. This statement is incorrect.
- B. Correct.
Residual connections help neural networks learn identity mappings, which facilitate better gradient flow during training and address issues like vanishing gradients. This is the correct benefit.
- C. Incorrect.
Residual connections do not skip layers entirely but instead add the input of a layer directly to its output, enabling deep networks to train effectively. This statement is incorrect.
- D. Incorrect.
Residual connections are specifically designed to bypass certain layers, not ensure sequential processing. This statement is incorrect.