Scribd
Upload
35 views

Assignments Module 10

The document outlines 3 assignments related to deep learning models: 1. Use the BERT model on a spam detection dataset, reproducing results and enhancing the epochs to observe any improvements in prediction accuracy. 2. Reproduce GAN results on MNIST to generate images and observe if the discriminator can be fooled, plotting a generated image at 4000 epochs. 3. Reproduce VAE results on Fashion MNIST for 100 and 50 epochs, observing any changes in latent space vectors or scatter plots between the two.

Uploaded by

mehtakinjalb
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
35 views

Assignments Module 10

The document outlines 3 assignments related to deep learning models: 1. Use the BERT model on a spam detection dataset, reproducing results and enhancing the epochs to observe any improvements in prediction accuracy. 2. Reproduce GAN results on MNIST to generate images and observe if the discriminator can be fooled, plotting a generated image at 4000 epochs. 3. Reproduce VAE results on Fashion MNIST for 100 and 50 epochs, observing any changes in latent space vectors or scatter plots between the two.

Uploaded by

mehtakinjalb
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Assignments on Module 10

1. Access the codes on BERT from the following GitHub link:

https://github.com/Sabyasachi123276/Deep-Learning-Tutorials/blob/main/BERT.ipynb

Use the dataset named 'spam.csv' [attached].

Here, the main objective is to create a system with the BERT model for classifying SMS
messages as spam or non-spam to enhance potential security. First, reproduce the results. Then,
enhance the 'epochs = 20' in the 'Training the model' and other relevant sections. Now, let us
know your observations. Do you see any improvements in the prediction accuracy?

2. Access the codes on GANs from the following GitHub link:


https://github.com/Sabyasachi123276/Deep-Learning-Tutorials/blob/main/GANs_example.ipynb

Reproduce the results to gain experience in how the generator is building the images with the
same qualities as the input MNIST dataset and trying to make the discriminator fool.

Now, plot the generated image at 4000 epoch. Let us know your following observations: Can we
generate the same quality of the input MNIST image? Will the discriminator be fooled by this
image quality?

MNIST dataset descriptions: The MNIST database contains 60,000 training images and 10,000
testing images of handwritten digits. The MNIST dataset is present in Keras; you do not have to
upload it externally while doing this assignment.

3. Access the codes on VAEs from the following GitHub link:

https://github.com/Sabyasachi123276/Deep-Learning-Tutorials/blob/main/VAEs_example.ipynb

First, reproduce the results for 100 epochs. Then reduce the epoch value to 50. State your
following observations:

(i) Do you observe any change in the training results according to their values in latent space
vectors in comparison to the results obtained in 100 epochs?

(ii) Do you observe any change in the scatter plot of training data on the basis of their values of
corresponding latent dimensions generated from the encoder in comparison to the results
obtained in 100 epochs?
Fashion MNIST dataset descriptions:

Fashion-MNIST is a dataset of Zalando's article images, which consists of a training set of


60,000 examples and a test set of 10,000 examples. Each example is a 28x28 grayscale image
associated with a label from 10 classes.

Each training and test example is assigned to one of the following labels:

● 0 T-shirt/top
● 1 Trouser
● 2 Pullover
● 3 Dress
● 4 Coat
● 5 Sandal
● 6 Shirt
● 7 Sneaker
● 8 Bag
● 9 Ankle boot

The Fashion MNIST dataset is present in Keras; you do not have to upload it externally while
doing this assignment.

You might also like