NCA-GENM Question 78
Select 3You are building a multimodal generative AI model that combines text and image inputs to generate descriptive captions for images. During the research phase, you notice that the model’s performance on certain datasets is significantly lower. Which factors should you consider to identify relationships or trends affecting the results?
- A
Imbalanced distribution of text and image samples in the dataset
- B
Bias in the text annotations provided in the dataset
- C
The size of the model’s embedding layers
- D
Variations in image resolution within the dataset
- E
The number of GPUs used during training
Show answer and explanation
Correct answers: A, B, D
Explanation
Identifying relationships and trends in a dataset requires examining factors like data distribution, potential biases, and variations in input formats. Imbalanced datasets, annotation bias, and image resolution inconsistencies can all significantly influence model performance. Hardware aspects like GPU count, while important for efficiency, do not directly affect such analyses.
- A. Correct.
Imbalanced distribution of text and image samples may lead to the model underperforming on underrepresented combinations, making it a key factor to analyze.
- B. Correct.
Bias in text annotations can lead to skewed model predictions, as the model might learn incorrect relationships between text and image pairs.
- C. Incorrect.
While embedding layer size affects model capabilities, it is not directly related to identifying relationships or trends in the data.
- D. Correct.
Variations in image resolution can cause inconsistencies in feature extraction, which may affect the model's ability to generalize effectively.
- E. Incorrect.
The number of GPUs used during training affects training speed but does not directly impact identifying trends or relationships in the dataset.