NVIDIA Generative AI Multimodal Sample Questions:

1. You are building a generative A1 model that combines text and image inputs to generate novel images. You have access to NVIDIA NeMo and want to leverage its pre-trained models and tools. Which NeMo modules or features would be MOST beneficial for this multimodal task? (Select all that apply)

A) NeMo's support for PyTorch Lightning for efficient training and scaling.
B) NeMo's core building blocks for constructing custom neural network architectures.
C) NeMo's TTS models for generating image descriptions.
D) NeMo's ASR models for processing text inputs.
E) NeMo's pre-trained language models for text understanding and feature extraction.

2. You are tasked with evaluating a multimodal A1 model that combines image and text inputs to generate product descriptions. You observe that the model performs well on common product categories (e.g., clothing, electronics) but struggles with niche categories (e.g., antique furniture, scientific instruments). Which of the following strategies would be MOST effective in improving the model's performance on niche categories?

A) Increase the overall size of the training dataset.
B) Decrease the learning rate during training.
C) Fine-tune the model on a dataset specifically curated for niche product categories.
D) Implement data augmentation techniques to create synthetic data for niche categories.
E) Replace the image encoder with a more powerful architecture.

3. You are working on a multimodal AI model that generates images from text descriptions. You notice that the generated images often lack fine-grained details and appear blurry. Which of the following techniques is LEAST likely to improve the visual quality of the generated images?

A) Reducing the batch size during training.
B) Employing a convolutional neural network (CNN) with strided convolutions for upsampling.
C) Training with a larger dataset of higher-resolution images.
D) Increasing the latent space dimensionality of the generative model.
E) Using a perceptual loss function that penalizes differences in high-level features.

4. Which of the following are valid techniques for fusing multimodal data?

A) Early Fusion
B) Intermediate Fusion
C) Late Fusion
D) Just Fusion
E) All of the above except D.

5. You are building a generative AI model that creates realistic product designs based on textual descriptions and a reference image depicting a similar, but not identical, product. You are using a Variational Autoencoder (VAE) architecture. However, the generated images lack the fine-grained details present in the reference image. Which of the following methods would be most suitable to incorporate fine-grained details from the reference image into the generated design?

A) Use a larger convolutional kernel Size in the decoder
B) Implement a skip connection from the encoder of the reference image to the decoder of the generative model, allowing the decoder to directly access features from the reference image at multiple scales.
C) Increase the latent space dimensionality of the VAE.
D) Replace the VAE with a Generative Adversarial Network (GAN).
E) Reduce the batch size during training.

Solutions: