NCA-GENM問題集、NVIDIA実際の試験問題

質問 1

You're developing a multimodal model that combines text and audio for sentiment analysis. The text component is performing well, but the audio component contributes very little to the overall accuracy. What's the MOST likely reason and how could you address it?

A. The audio data is not preprocessed correctly. Apply aggressive noise reduction techniques.

B. The audio features are not properly aligned with the text features. Use a cross-modal attention mechanism to improve alignment.

C. The audio data is irrelevant. Remove the audio component entirely.

D. The text component is simply too dominant. Reduce the weight given to the text component in the final prediction.

E. The audio data is too large. Downsample the audio data to reduce computational cost.

正解: B

解説: (PassTest メンバーにのみ表示されます)

質問 2

You are training a deep convolutional generative adversarial network (DCGAN) for generating high-resolution images. After several epochs, you observe mode collapse the generator produces only a few similar images. Which of the following strategies would be most effective in mitigating mode collapse?

A. Decrease the learning rate of the generator and discriminator simultaneously.

B. Increase the batch size significantly to provide the discriminator with a more diverse set of samples.

C. Introduce batch normalization only in the generator network.

D. Use label smoothing in the discriminator to penalize overconfident predictions.

E. Implement feature matching in the discriminator by making the generator learn to match intermediate layer activations of the discriminator on real data.

正解: E

解説: (PassTest メンバーにのみ表示されます)

質問 3

Which of the following techniques is MOST suitable for aligning the feature spaces of text and images in a multimodal model?

A. Employing a contrastive loss function that encourages similar representations for semantically related text and images.

B. Training the text and image encoders independently.

C. Using separate loss functions for text and image encoders.

D. Only using image data during the training process.

E. Concatenating the features from the text and image encoders without any further processing.

正解: A

解説: (PassTest メンバーにのみ表示されます)

質問 4

You are fine-tuning a pre-trained large language model (LLM) for a specific text generation task. During training, you observe that the model is overfitting to the training data and not generalizing well to unseen examples. Which of the following techniques could be MOST effective in mitigating overfitting in this scenario?

A. Increasing the size of the training dataset.

B. Early stopping based on a validation set.

C. Using a smaller batch size during fine-tuning.

D. Decreasing the learning rate during fine-tuning.

E. Applying dropout regularization to the LLM's layers.

正解: B,D,E

解説: (PassTest メンバーにのみ表示されます)

質問 5

You are experimenting with different multimodal transformer architectures for a video understanding task. You are using a large pre- trained model and fine-tuning it on your specific dataset. You observe that the model is overfitting and struggling to generalize to unseen videos. Which of the following techniques would be most effective in mitigating overfitting in this scenario? (Choose two)

A. Employ data augmentation techniques specifically designed for video data (e.g., temporal jittering, random cropping).

B. Reduce the number of transformer layers in the model.

C. Use a smaller pre-trained model.

D. Increase the batch size significantly.

E. Implement weight decay and dropout regularization.

正解: A,E

解説: (PassTest メンバーにのみ表示されます)

質問 6

When deploying a Generative A1 model to a resource-constrained edge device (e.g., a mobile phone), what are the key considerations for model optimization and which techniques are most effective?

A. A and B.

B. Knowledge distillation to transfer knowledge from a large, accurate model to a smaller, faster model suitable for the edge.

C. Increasing the model's complexity to improve accuracy on the limited hardware.

D. Focusing solely on reducing the model's parameter count without considering its computational complexity.

E. Model quantization (e.g., INT8) to reduce model size and memory bandwidth, and model pruning to remove unimportant connections.

正解: A

解説: (PassTest メンバーにのみ表示されます)

質問 7

You are developing a system that uses multimodal data (images, audio, and text) to detect fraudulent insurance claims. The image data represents damage to vehicles, the audio data captures conversations between the claimant and the insurance agent, and the text data includes the claim form details. What are the potential benefits of using multimodal data compared to relying on a single modality?

A. Improved accuracy and robustness due to complementary information from different modalities.

B. Increased vulnerability to adversarial attacks and data noise.

C. Reduced computational complexity and training time compared to using a single modality.

D. Simplified data preprocessing and feature engineering.

E. Ability to handle missing or incomplete data in one modality by relying on information from other modalities.

正解: A,E

解説: (PassTest メンバーにのみ表示されます)

質問 8

You are building a system that generates image captions from images and vice vers a. Which evaluation metric(s) are MOST appropriate to assess the quality of the generated content? (Select all that apply)

A. Inception Score

B. ROUGE score

C. Accuracy

D. FID (Frechet Inception Distance)

E. BLEU score

正解: B,D,E

解説: (PassTest メンバーにのみ表示されます)

質問 9

You are working on a multimodal emotion recognition system that analyzes video (visual and audio) and transcript (text) dat a. You want to fuse these modalities effectively. Which fusion technique is MOST likely to capture complex inter-modal relationships and improve performance, especially when the modalities have varying degrees of reliability?

A. Late fusion (averaging the probabilities from separate modality-specific models).

B. Early fusion (concatenating features before feeding into a single model).

C. Feature-level averaging.

D. Attention-based fusion (using attention mechanisms to weigh the contributions of each modality dynamically).

E. Simple concatenation of modality-specific embeddings at a single point in the model.

正解: D

解説: (PassTest メンバーにのみ表示されます)

質問 10

You are building a multimodal model for medical image diagnosis, using both radiology images (e.g., X-rays) and patient clinical notes.
The clinical notes are highly unstructured and contain significant medical jargon. What preprocessing steps would be MOST effective for improving the model's performance?

A. Performing sentiment analysis on the clinical notes.

B. Translating the clinical notes into multiple languages and then back-translating to the original language.

C. Utilizing named entity recognition (NER) to identify medical entities (diseases, medications, etc.), and employing a medical-specific language model (e.g., BioBERT) for text embeddings.

D. Applying basic text cleaning (removing punctuation, converting to lowercase) and using a standard word embedding (e.g., Word2Vec).

E. Directly feeding the raw clinical notes into the model without any preprocessing.

正解: C

解説: (PassTest メンバーにのみ表示されます)

質問 11

A multimodal A1 model is designed to translate sign language videos into text. The model performs well on videos with clear hand gestures and lighting conditions but struggles with videos recorded in low light or with partial hand occlusions. Which of the following strategies would be MOST effective in improving the model's robustness to these challenging conditions?

A. Using a simpler text encoder.

B. Training the model on a smaller dataset.

C. Reducing the frame rate of the input videos.

D. Increasing the size of the text vocabulary.

E. Applying image enhancement techniques (e.g., contrast adjustment, noise reduction) to the video frames.

正解: E

解説: (PassTest メンバーにのみ表示されます)

質問 12

Consider a multimodal emotion recognition system that uses both facial expressions (images) and speech (audio). You want to fuse the information from these two modalities at the decision level. Which of the following techniques would be MOST suitable for decision-level fusion?

A. Train separate classifiers for images and audio, then use a weighted average of their output probabilities based on the confidence scores of each classifier.

B. Train separate classifiers for images and audio, then use the output of the image classifier as input to the audio classifier-

C. Train separate classifiers for images and audio, then average their output probabilities for each emotion class.

D. Concatenate the feature vectors extracted from the images and audio, then train a single classifier.

E. Train a single transformer to process both images and audio in sequence.

正解: A

解説: (PassTest メンバーにのみ表示されます)

質問 13

Consider the following Python code snippet using PyTorch. What does this code do in the context of data preprocessing for a Generative AI model?

A.

B.

C.

D.

E.

正解: A

解説: (PassTest メンバーにのみ表示されます)

質問 14

You are building a multimodal model that takes images and text descriptions as input to generate new images. You want to evaluate the impact of different image encoders (ResNet50, Efficient Net) on the generated image quality and relevance to the text prompt. Which evaluation metric(s) would be MOST appropriate for this task?

A. Inception Score (IS) only

B. CLIP Score only

C. Frechet Inception Distance (FID) only

D. Perplexity and BLEU score

E. Both FID and CLIP Score

正解: E

解説: (PassTest メンバーにのみ表示されます)

NVIDIA Generative AI Multimodal - NCA-GENM 模擬練習