[go: up one dir, main page]
Scribd
Upload
111 views2 pages

Neural Machine Translation - Coursera

Uploaded by

Priyanka
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
111 views2 pages

Neural Machine Translation - Coursera

Uploaded by

Priyanka
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

1.Which of the following are bottlenecks when implementing seq2seq models?

1 / 1 point
You are trying to store variable length sequences in a fixed memory, for example, you are trying to store articles
of different lengths in a fixed 100 dimensional vector.
Correct
Correct
There are vanishing/exploding gradient problems.

Correct
Correct
They require a lot of memory.

They can directly handle sequences of variable lengths without any modification.

2.What are some of the benefits of using attention? 1 / 1 point


It helps with the information bottleneck issue.

Correct
Correct.
It improves the interpretability of the model by highlighting which parts of the input contribute to the output.

Correct
Correct
It allows you to focus on the parts that matter more.

Correct
Correct.
It simplifies the model architecture by reducing the need for complex recurrent layers.

Correct
Correct.
3.In the context of Transformer models, which components are essential for the attention mechanism? Select all that 1 / 1 point
apply.
Keys: Described as part of the mechanism that helps in identifying relevant information from the input data.

Correct
Keys help the model pick out the relevant bits from your data, matching them with queries.
Cosine similarity: While useful in measuring similarity between vectors, it is not an essential component of the
attention mechanism in Transformer models.
Queries: Described as the component that represents the current item being processed, to find matching
information.
Correct
Queries act like a spotlight, focusing on what's currently important in the model's task.
Values: These hold the actual information from the input that will be used to compute the output, once a match
is found between a key and a query.
Correct
Once the model finds a match between keys and queries, values provide the actual information we want to
use.
Activation Functions: While activation functions like ReLU are crucial in neural network layers for introducing
non-linearity, they are not considered an essential component of the attention mechanism itself. (Incorrect) -
This replaces the previously correct option.
4.Teacher forcing uses the actual output from the training dataset at time step y (t) as input in the next time step X (t+1) , 1 / 1 point
instead of the output generated by your model.

True.

False.

Correct
Correct.
5.The BLEU score's range is as follows: 1 / 1 point
The closer to 0, the worse it is, the closer to 1, the better it is.

The closer to 1, the worse it is, the closer to 0, the better it is.
The closer to -1, the worse it is, the closer to 1, the better it is.

The closer to −∞, the worse it is, the closer to ∞, the better it is.

Correct
Correct.
6.BLEU (Vanilla Implementation) is defined as: 1 / 1 point
(Sum of unique n-gram counts, overlapping in the candidate and reference) / (Total # of n-grams in the
candidate)
(Sum of unique n-gram counts in the candidate) / (Total # of n-grams in the candidate)

(Sum of unique unigram counts in the candidate) / (Total # of n-grams in the reference)

(Sum of unique unigram counts, overlapping in the candidate and reference) / (Total # of unigrams in the
reference)
Correct
Correct.
7.What aspect of text summaries does the Rouge metric primarily evaluate? 1 / 1 point
The similarity of the summary to reference summaries.

The grammatical correctness of the summary.

The uniqueness of words in the summary.

Rouge doesn't evaluate the summary based on its length. It's more about comparing the content to reference
summaries.
Correct
Rouge measures how well the summary captures the key points found in reference summaries. It's all about
matching content.
8.Greedy decoding 1 / 1 point
Allows you select the word with the highest probability at each time step.

Allows you randomly select the word according to its own probability in the softmax layer.

Selects multiple options for the best input based on conditional probability.

Makes use of the Minimum Bayes Risk method.

Correct
Correct.
9.When implementing Minimum Bayes Risk method in decoding, let's say with 4 samples, you have to implement the
following.
1. Calculate similarity score between sample 1 and sample 2
2. Calculate similarity score between sample 1 and sample 3
3. Calculate similarity score between sample 1 and sample 4
4. Average the score of the first 3 steps (Usually a weighted average)
5. Repeat until all samples have overall scores

Pick the best candidate, with the highest similarity score.


False

True

Correct
Correct.

You might also like