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.Increase the latent space dimensionality of the VAE.
• B.Use a larger convolutional kernel Size in the decoder
• C.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.
• D.Reduce the batch size during training.
• E.Replace the VAE with a Generative Adversarial Network (GAN).

Comment: *

Name: *

Email: *

Verification: *

What is a main application of Triton Inference Server?

• A.Triton Server can be used to generate images from pure noise.
• B.Triton Server can be used to deploy AI models on the GPU only.
• C.Triton Server can be used to execute GPU-accelerated graph analysis with cuGraph.
• D.Triton Server can be used to deploy neural networks from various frameworks.

Comment: *

Name: *

Email: *

Verification: *

You have developed a multimodal model that uses both audio and video data to detect human emotions. During testing, you observe that the model performs exceptionally well on controlled lab recordings but poorly in real-world scenarios with background noise and varying lighting conditions. What technique would be MOST effective in improving the model's generalization ability to real-world data?

• A.Reducing the model's complexity to prevent overfitting to the lab recordings.
• B.Increasing the amount of data from lab recordings.
• C.Data augmentation techniques such as adding noise to the audio, simulating different lighting conditions for the video, and using transfer learning from pre- trained audio and video models.
• D.Training separate models for lab recordings and real-world data.
• E.Replacing the audio input with text transcripts.

Comment: *

Name: *

Email: *

Verification: *

You are developing a multimodal system for medical diagnosis that integrates patient history (text), X-ray images, and heart rate data (time-series). A significant portion of the heart rate data is missing due to sensor failures. What is the MOST appropriate method to handle this missing data to ensure the model's accuracy and prevent bias?

• A.Replace the missing heart rate data with the mean heart rate value calculated from the available data.
• B.Remove the records of patients with missing heart rate data from the dataset.
• C.Use a time-series imputation technique, such as Kalman filtering or recurrent neural networks, to estimate the missing heart rate values based on the available data and temporal patterns.
• D.Assign a fixed, arbitrary value (e.g., 0) to all missing heart rate data points.
• E.Train a separate model specifically on data without the time-series component.

Comment: *

Name: *

Email: *

Verification: *

You are building a multimodal model to predict stock prices using financial news articles (text), historical stock prices (time-series), and company logos (images). You have preprocessed the data and are ready to train your model. Which of the following architectures would be MOST suitable for effectively integrating these three modalities?

• A.A simple feed forward neural network with concatenated features from all modalities.
• B.A model that combines a Transformer for text, an LSTM for time-series, and a CNN for images, with a late fusion strategy using a weighted averaging of predictions.
• C.A model that uses a Transformer encoder for each modality, followed by a shared Transformer decoder for prediction, enabling cross-modal attention at the decoder level.
• D.Separate models for each modality trained independently, and then ensembled together at the prediction stage.
• E.A model that converts all data into a single text format and uses a large language model (LLM) for prediction.

Comment: *

Name: *

Email: *

Verification: *