Free NVIDIA Generative AI Multimodal Practice Questions

A.  

A/B testing is used to measure the impact of changes in the user interface of a ML application.

B.  

A/B testing helps in optimizing the hyperparameters of a machine learning model.

C.  

A/B testing is irrelevant in ML software engineering.

D.  

A/B testing helps in evaluating the performance and effectiveness of different machine learning models.

A.  

Decision Trees

B.  

K-Means Clustering

C.  

Convolutional Neural Networks (CNN)

D.  

Linear Regression

A.  

To generate new data based on the training set.

B.  

To distinguish between real and generated data.

C.  

To optimize the training process.

D.  

To calculate the loss function and update the generator.

A.  

The data used to train the AI system may not be representative of the population it is intended to serve.

B.  

Bias in data is irrelevant as long as the AI system produces accurate predictions.

C.  

Bias in data can only be addressed after the AI system has been deployed.

D.  

Bias in data is not a concern for AI systems as they are designed to be neutral and objective.

A.  

State management and composition

B.  

Transfer learning

C.  

Prompt engineering

D.  

Neural network integration

A.  

Unimodal models have a higher risk of gradient vanishing compared to multimodal models, as the focus on a single modality allows for better gradient flow and stability.

B.  

Multimodal models have a higher risk of gradient vanishing compared to unimodal models, as the combination of multiple modalities increases the complexity of the model architecture.

C.  

Both multimodal and unimodal models have an equal risk of gradient vanishing, as the optimization process is independent of the number of modalities.

D.  

Gradient vanishing is not a concern in either multimodal or unimodal models, as modern optimization techniques have overcome this issue.

A.  

CTC Loss

B.  

F1 Score

C.  

Mean Opinion Score (MOS)

D.  

Word Error Rate (WER)

A.  

The batch size has no impact on VRAM consumption during inference.

B.  

Increasing or decreasing the batch size has the same impact on VRAM consumption.

C.  

Increasing the batch size reduces VRAM consumption because more data can be processed in parallel.

D.  

Decreasing the batch size reduces VRAM consumption.

A.  

Padding in convolution increases the spatial dimensions of the input feature map, while padding in transposed convolution decreases the spatial dimensions of the output feature maps.

B.  

In a convolution operation, padding is added to the output after it has been expanded with the stride. On the other hand, in a transposed convolution operation, padding is added to the input before it is expanded with stride.

C.  

Padding in convolution enables convolution operations on the boundary pixels of the input. In transposed convolution, it removes rows and columns along the perimeter of the input after it is expanded with stride.

D.  

Padding in convolution and transposed convolution serve the same purpose of reducing the convolutional neural network's memory requirement and computational cost of the convolutional neural network.

E.  

Padding in convolution is used only when the input image is smaller than the filter size, while padding in transposed convolution is used only when the input image is larger than the filter size.

A.  

Pruning weights or neurons.

B.  

Adding more convolutional filters.

C.  

By replacing activation functions in some neurons with simpler ones.

D.  

Increasing the batch size.