Reference no: EM133482217

Question 1: Explain what is Transfer Learning concept with an application of how Transfer Learning implemented as example.

Based on your example, discuss the advantages and disadvantages in using Transfer Learning. (5 marks)

Question 2

Suppose you have been approached for your deep learning expertise. You have to classify images of road with 3 different weather.

Left top and bottom images rainy (class label = 0)
Middle top and bottom images foggy (class label = 1)
Right top and bottom images Snowy (class label = 2)
You notice that the training set only comprises images captured during the day, whereas the testing set only contains pictures shot at night after visually analysing the dataset. Describe the problem and how you would solve it (10 marks).
You also discover you do not have enough data when you train your model. Provide one data augmentation approaches with details that can be employed to overcome data limitation (10 marks).

Question 3

Assume a 5-node input layer, 3-node hidden layer, and 5-node output layer for an autoencoder.

Draw the autoencoder with above given nodes and show connections in between them. Name each layers, and indicate encoder and decoder parts clearly.
Given 4-dimensional dataset below:
A B C D
1 0 0 0
0 1 1 1
1 0 0 0
1 0 1 0
0 0 1 0
Answer the following questions to describe how the autoencoder above is trained:

How will the autoencoder be supplied with input and output values? Give an example to support your response (5 marks).
When will the training come to end? Make a list of at least two factors for stopping. (5marks)

Section 2

Assume you are working for an energy company, and you were assigned a task to create a model to predict the total electricity consumed in Victoria.

You were provided a dataset with the daily consume electricity in Victoria for 12,053 days (about 396 months) from 1st Jan 1990 to 31st Dec 2022.

Your first task will be predicting the total electricity consumed in January 2023 in Victoria.

Please answer the following questions:

Question 4

Can CNN be used in this scenario? Justify your answer

Question 5

Discuss the best way to use the dataset, and discuss whether there is any preprocess to the dataset needs to be conducted? Justify your answer.

Question 6

Design an architecture of network to train the model based on your answer in Q4. and explain how the model will be trained and why the total electricity consumed in January can be predicted.

Question 7

Is there any potential data privacy issue or professional issue in this scenario? Justify your answer.