NVIDIA Generative AI Multimodal Sample Questions:

1. You are tasked with building a system that generates realistic images from text descriptions. Which of the following loss functions is MOST crucial for ensuring the generated images are both visually appealing and semantically aligned with the text?

A) All of the above.
B) A perceptual loss that measures the difference in high-level features extracted by a pre-trained convolutional neural network.
C) A contrastive loss that encourages similar text and image embeddings to be closer together in a shared embedding space.
D) Binary Cross-Entropy loss for discriminating between real and generated images (used in GANs).
E) Mean Squared Error (MSE) between the generated image and the target image (if available).

2. You're building a multimodal model that takes images and text as input. You notice that your model is heavily biased towards the text modality, essentially ignoring the visual input. Which of the following strategies could you employ to address this modality imbalance? (Select TWO)

A) Remove the text modality entirely.
B) Increase the learning rate for the text encoder.
C) Reduce the size of the visual encoder.
D) Implement a gating mechanism that dynamically adjusts the contribution of each modality based on the input.
E) Use a modality-specific loss function, weighting the loss from the visual modality more heavily.

3. You are tasked with integrating a CLIP model into your application to generate images based on text descriptions. You want to ensure that the generated images closely reflect the nuances of the text prompt. Which prompt engineering technique is MOST suitable for achieving this?

A) Using short, concise prompts to minimize ambiguity.
B) Using prompts consisting only of keywords related to the desired image.
C) Using overly verbose and descriptive prompts to maximize detail.
D) Using negative prompts to explicitly exclude unwanted features or styles.
E) Using random prompts to explore the model's creative capabilities.

4. When using prompt engineering with text-to-image models, which of the following techniques are most effective in improving the fidelity and relevance of generated images to the input text?

A) Using negative prompts to explicitly exclude undesirable elements from the generated image.
B) Using highly specific and detailed prompts, including attributes, style, and composition.
C) Using vague and open-ended prompts to encourage creative variations.
D) Focusing solely on the main subject of the image, omitting any contextual details.
E) Using a combination of highly specific prompts and negative prompts.

5. You are experimenting with different architectures for a text-to-speech (TTS) model. You have implemented a Tacotron 2 model and a FastSpeech 2 model. Which of the following statements accurately describes the key differences between these two architectures and their implications?

A) Both Tacotron 2 and FastSpeech 2 use attention mechanisms, but FastSpeech 2 incorporates length regulator and variance adaptor modules to address the one-to-many mapping problem, leading to more stable and controllable synthesis.
B) FastSpeech 2 uses an attention mechanism for aligning text and speech, while Tacotron 2 relies on a fixed alignment, resulting in faster training and inference for Tacotron 2 but potentially lower quality.
C) Both Tacotron 2 and FastSpeech 2 rely on fixed alignments, but FastSpeech 2 uses a more complex decoder architecture, leading to higher quality but slower inference.
D) Tacotron 2 is an autoregressive model, while FastSpeech 2 is a non-autoregressive model. This allows FastSpeech 2 to generate speech in parallel, resulting in significantly faster inference speeds.
E) Tacotron 2 uses an attention mechanism for aligning text and speech, while FastSpeech 2 relies on a fixed alignment, resulting in faster training and inference for FastSpeech 2 but potentially lower quality.

Solutions: