Generate Fake

what is a target:

$($b$)$ pts$)$ True or

${40}^{**}90-G$

S$22\_$final.pdf

vent of Code

Leetcode

ChatGPT

jupyter

Perplexity

$72\%$

$2.$ Single or Multiple Choices $(30$ points$)$

$($a$)(3$ pts$)$ Which of the followings is NOT an unsupervised learning method?

A$.$ Principal component analysis.

B$.$ Autoencoder.

C$.$ K$-$means clustering.

D$.$ Polynomial regression.

$($b$)(3$ pts$)$ We want to train a recurrent neural network on a small dataset for sentiment analysis. Which of the followings helps alleviate overfitting?

A$.$ Self$-$attention.

B$.$ Stacked LSTM$.$

C$.$ Bidirectional LSTM$.$

D$.$ Pretrain the embedding layer on a large dataset.

$($c$)(3$ pts$)$ The input is a $20153$ tensor. What is the number of $543$ patches in the input tensor? Here we set stride $=1$ and no zero$-$padding.

A$.11.$

B$.43.$

C$.54.$

D$.1511.$

E$.1612.$

F$.2015$

$($d$)(3$ pts$)$ We want to learn a sequence$-$to$-$sequence model for machine translation. We hope the model will be less likely to "forget". Which of the following approaches is helpful?

A$.$ In the decoder network, use bidirectional LSTM instead of LSTM$.$

B$.$ In the encoder network, use bidirectional LSTM instead of LSTM$.$

C$.$ In the encoder network, use stacked LSTM instead of LSTM$.$

D$.$ In the decoder network, use stacked LSTM instead of LSTM$.$

$($e$)(3$ pts$)$ For training denoising autoencoder to remove noise from images. The training data should be:

A$.$ Clean images as inputs, and clean images as targets.

B$.$ Noisy images as inputs, and noisy images as targets.

C$.$ Clean images as inputs, and noisy images as targets.

D$.$ Noisy images as inputs, and clean images as targets.

Page $3$