Crowdly
Додати до Chrome
Deep Learning (2024/2025)
Шукаєте відповіді та рішення тестів для Deep Learning (2024/2025)? Перегляньте нашу велику колекцію перевірених відповідей для Deep Learning (2024/2025) в moodle.kent.ac.uk.
Отримайте миттєвий доступ до точних відповідей та детальних пояснень для питань вашого курсу. Наша платформа, створена спільнотою, допомагає студентам досягати успіху!
You want to train a simple RNN next-character prediction model on the sample text:
The model should predict the most likely character given a sequence of previous characters:
- Show the process of data preparation step by step and demonstrate how to encode it using one-hot encoding. Ensure that all steps are clearly shown. Half marks will be awarded if the workings are not provided. [20 marks]
- Using a window size of 5, generate and illustrate your training and target (labels) datasets for model training. Display only the first and last elements of each. What are the shapes of the x_train and y_train? [15 marks]
- Comment on using the categorical_crossentropy as your loss function for model training. Is ReLU a good activation function for this problem and why? [5 marks]
Note! It is acceptable to use shortcuts for displaying answers. For example, instead of showing the full vector like [0 0 0 0 0 0 0 0 0 0 0 0], you can represent it as [0 0 0 .... 0], and indicate that the vector has 12 elements of 0s.
You are given a Convolutional Neural Network with an input of 4x4 on the left-hand side table and a kernel of 3x3 on the right-hand side table below.
4
|
8
|
0
|
1
|
0
|
0
|
2
|
4
|
0
|
1
|
5
|
3
|
7
|
1
|
0
|
0
|
Input
1
|
0
|
-1
|
0
|
0
|
0
|
-1
|
0
|
1
|
Kernel
Part A:
1. Apply padding = 1. Draw the input with padding (table style) and explain what you did
2. Using a stride = 1, padding = 1 of the input table, and the given kernel, calculate the output of the convolutional layer, showing how you computed each output pixel. Draw the final output table ( the output table must be of the same dimension as the input table) [25 marks].
Part B:
3. Using the original input table, down-sample the input using using a 2x2 filter. Draw the result of the pooling layer and show how you computed each pixel, explaining what the average pooling operation does
Suppose you have a dataset of 30,000 images.
a) Draw a schematic diagram to illustrate how you would train and test your custom deep learning model using k-fold cross-validation with k = 5. Use visual representation to indicate how you split the data
b) Show how you numerically calculate the total average performance of the model across all folds (i.e., summarise the model’s performance). [20 marks]