NCA-GENM Question 23
Single answerYou are tasked with designing a neural network for a multimodal generative AI application. The network must efficiently learn complex relationships between input data modalities while avoiding vanishing gradient issues during training. Which of the following design choices would be most effective for achieving this goal?
- A
Use a nonsequential neural network with residual connections
- B
Implement a fully sequential neural network with no skip connections
- C
Increase the depth of the network without modifying the architecture
- D
Replace all activation functions with linear activation
Show answer and explanation
Correct answer: A
Explanation
Residual connections, a key feature of nonsequential neural networks, are designed to address the vanishing gradient problem by allowing gradients to flow more effectively through the network. This enables deeper architectures to learn efficiently and capture complex relationships, making them highly suitable for multimodal generative AI applications.
- A. Correct.
Residual connections in a nonsequential neural network allow information to bypass certain layers, mitigating vanishing gradient issues and enabling the network to learn complex relationships more effectively.
- B. Incorrect.
A fully sequential neural network without skip connections is prone to vanishing gradient problems, especially as the network becomes deeper, making it less suitable for the given scenario.
- C. Incorrect.
Merely increasing the depth of the network without addressing the vanishing gradient problem would likely exacerbate training difficulties and reduce performance.
- D. Incorrect.
Replacing all activation functions with linear activation would severely limit the network's ability to model non-linear relationships, which are critical for understanding complex data modalities.