Question 6: Deep Learning
a. Given the following image (I) and filter (F), compute the resulting feature map when using a step of two. (4 marks)
Image (I)
0 | 1 | 0 | 1 |
1 | 2 | 0 | -1 |
0 | 1 | 1 | 2 |
3 | 1 | 1 | 0 |
Filter (F)
3 | 2 |
1 | 2 |
b. You are in charge of developing a machine learning system to predict the number of students that will take a campus accommodation next year. For this, you have data from the past five years containing: “Number of students enrolled”, “Number of accommodation places available” and the “Number of students that booked campus accommodation”. Can you use a deep learning approach for this system? Justify your choice. (2 marks)
c. A biologist contacts you for your deep learning expertise. They want to classify images of plants into 100 classes, each class for a species of at-risk plant. For this, they have already collected a dataset of 1000 images which you split in 50% training and 50% testing. Using the below neural network architecture you get on the training set an accuracy of 99%, but on the testing set the accuracy is of 60%. What options can you try in order to improve the testing set accuracy?
(4 marks)
