AML AML Natural Language Processing & Computer Vision 2 — Questions and Answers
Question 1: What is transfer learning in the context of computer vision?
- Training a model from scratch on a new dataset
- Using a pre-trained model's weights as a starting point for a new task (Correct answer)
- Transferring labels from one dataset to another
- Copying model architecture without pre-trained weights
Correct answer: Using a pre-trained model's weights as a starting point for a new task
Transfer learning reuses the feature representations learned from a large dataset (e.g., ImageNet) to improve performance on a smaller target task.
Question 2: Which technique is used to augment training images by randomly flipping, rotating, or cropping them?
- Dropout regularization
- Batch normalization
- Data augmentation (Correct answer)
- Weight decay
Correct answer: Data augmentation
Data augmentation artificially expands the training set with label-preserving transformations, reducing overfitting in vision models.
Question 3: In NLP, what is tokenization?
- Converting words to their base grammatical form
- Splitting raw text into individual units such as words or subwords (Correct answer)
- Mapping tokens to their embedding vectors
- Removing stop words from a sentence
Correct answer: Splitting raw text into individual units such as words or subwords
Tokenization breaks raw text into discrete units (tokens) such as words, subwords, or characters that a model can process.
Question 4: What is the attention mechanism in transformer models primarily designed to do?
- Reduce gradient vanishing during backpropagation
- Allow the model to weigh the relevance of different input tokens when producing each output (Correct answer)
- Increase the batch size during training
- Replace the need for positional encodings
Correct answer: Allow the model to weigh the relevance of different input tokens when producing each output
Attention computes a weighted sum of value vectors, where weights reflect how relevant each input token is to the current output position.
Question 5: Which metric is commonly used to evaluate object detection models by measuring overlap between predicted and ground-truth bounding boxes?
- F1 Score
- BLEU Score
- Intersection over Union (IoU) (Correct answer)
- ROUGE Score
Correct answer: Intersection over Union (IoU)
IoU divides the area of overlap between the predicted and actual bounding boxes by their combined union area, ranging from 0 to 1.
Question 6: What is the BLEU score used to measure in NLP?
- Classification accuracy on text labels
- Quality of machine-generated text by comparing n-gram overlap with reference translations (Correct answer)
- Semantic similarity between two sentence embeddings
- Number of out-of-vocabulary tokens in a corpus
Correct answer: Quality of machine-generated text by comparing n-gram overlap with reference translations
BLEU measures how many n-grams in a machine translation match those in one or more reference translations, normalized by length.
AML AML Natural Language Processing & Computer Vision 2
What is transfer learning in the context of computer vision?