NCA-GENM Question 85
Select 3You are assisting in the development of a multimodal AI model that combines text and image inputs. During testing, the model's accuracy for generating captions from images is significantly lower than expected. Which of the following steps would most likely help identify or resolve the issue?
- A
Check if the training dataset contains a balanced variety of image-text pairs.
- B
Fine-tune the model on a pre-trained text-only dataset to improve text generation quality.
- C
Evaluate the performance of each individual modality (text and image) before combining them.
- D
Inspect the preprocessing pipeline to ensure consistent formatting of image and text inputs.
- E
Adjust the model architecture to reduce the number of layers in the text encoder.
Show answer and explanation
Correct answers: A, C, D
Explanation
When developing and testing multimodal AI models, it is essential to ensure the dataset is representative, evaluate each modality separately to identify specific weaknesses, and verify that the preprocessing pipeline handles inputs correctly. These steps help isolate and address issues affecting model performance.
- A. Correct.
Checking the training dataset for balance and variety ensures that the model is exposed to diverse examples, reducing biases and improving performance.
- B. Incorrect.
Fine-tuning on a text-only dataset may improve general text generation but does not address the multimodal nature of the problem or the specific image-captioning issue.
- C. Correct.
Evaluating each modality independently helps isolate whether the issue lies with the text processing, image processing, or their integration.
- D. Correct.
Inspecting the preprocessing pipeline ensures there are no inconsistencies or errors in how inputs are formatted, which can directly impact model performance.
- E. Incorrect.
Reducing the number of layers in the text encoder is unlikely to resolve the issue and may degrade performance by limiting the model's ability to understand complex text representations.