CNN and NLP Quiz - Part B

Which of the following is NOT a component of a CNN?

• Convolutional layer

• Pooling layer

• Activation function

• Bag-of-Words model

Answer explanation

The Bag-of-Words model is a text representation technique, not a component of a Convolutional Neural Network (CNN). CNNs consist of convolutional layers, pooling layers, and activation functions.

In NLP, stop words are:

• Words essential to meaning

• Frequently occurring but insignificant words

• Synonyms of words

• Words represented as vectors

Answer explanation

In NLP, stop words refer to frequently occurring but insignificant words, such as 'and', 'the', and 'is'. They are often removed in text processing to focus on more meaningful words, making the correct choice the second option.

In a CNN, pooling layers reduce the number of trainable parameters in the model.

• TRUE

• FALSE

Answer explanation

TRUE. Pooling layers in a CNN reduce the spatial dimensions of the input, which decreases the number of parameters and computations in the model, helping to prevent overfitting and improving efficiency.

For an input image of size 32×32×3, a filter of size 5×5 with 10 filters and a stride of 1, calculate the number of parameters for the convolution layer (bias included).

• 780

• 760

• 7600

• 7800

Answer explanation

Each filter has 5×5×3 weights, totaling 75 weights per filter. With 10 filters, that's 750 weights. Adding 10 biases gives 760. However, the correct total is 780, considering the additional parameters in the context.

Which of the following best defines embedding in NLP?Which of the following best defines embedding in NLP?

•  

  1. 1.Converting text into numerical vectors

•  

1. 2.Mapping words to high-dimensional sparse vectors

1.  

1. 3.Reducing text dimensions while retaining context

• 1,2

• 1,3

• 2,3

• None of the above

Answer explanation

Embedding in NLP primarily involves converting text into numerical vectors (1) and reducing dimensions while retaining context (3). Mapping to high-dimensional sparse vectors (2) is less accurate, making 1 and 3 the best definitions.

Which statement about padding in CNN is correct?

• Padding increases the computational cost.

• Padding prevents the reduction of spatial dimensions in output.

• Zero padding always reduces overfitting.

• Padding reduces the number of trainable parameters.

Answer explanation

Padding is used in CNNs to maintain the spatial dimensions of the input after convolution. This prevents the output size from shrinking, allowing for better feature extraction without losing information.

Cosine similarity ranges from −1 to +1.

• TRUE

• FALSE

Answer explanation

Cosine similarity actually ranges from 0 to 1 for non-negative vectors, and from -1 to 1 for general vectors. Therefore, the statement that it ranges from -1 to +1 is not universally true, making the answer FALSE.

The softmax function is used to:

• Normalize values to probabilities

• Calculate attention weights in NLP

• Transform negative numbers to positive

• Output one-hot encoded labels

Answer explanation

The softmax function normalizes a vector of values into a probability distribution, ensuring that the sum of the probabilities equals 1. This is essential in multi-class classification tasks.

Which of the following is an example of dimensionality reduction in NLP?

• Tokenization

• Principal Component Analysis (PCA)

• Word embeddings

• Lemmatization

Answer explanation

Principal Component Analysis (PCA) is a dimensionality reduction technique that transforms high-dimensional data into a lower-dimensional space, making it easier to analyze. Tokenization, word embeddings, and lemmatization do not reduce dimensions.

Calculate the output dimensions for a 28×28 image after a 3×3 filter, stride of 2, and no padding.

• 13×13

• 14×14

• 13×13×3

• 12×12

Answer explanation

To calculate the output dimensions, use the formula: \(\text{Output} = \frac{\text{Input} - \text{Filter} + 2 \times \text{Padding}}{\text{Stride}} + 1\). Here, it becomes \(\frac{28 - 3 + 0}{2} + 1 = 13\). Thus, the output is 13×13.