This module looks into the training of modern deep neural networks: backpropagation, regularisation, automatic differentiation, computational graphs. Introduces different types of deep neural networks, such as, LSTM, convolutional networks, and autoencoders. Presents the theoretical underpinnings of deep learning and its mechanisms. Delves into selected recent advanced topics in deep learning. Examines applications of deep learning.
Private Study Hours: 124
Contact Hours: 26
Total Hours: 150
This module will be assessed by 100% coursework.
50% Practical assignment 1 (individual; approximately 40 hours), with an additional challenging task with respect to COMP6685 assessment 1
25% Demo on practical assignment 1 (individual; 1 hour with 10 hours revision)
25% Time limited assessment 2 (individual; 1 hour with 10 hours revision), with a different set of questions with respect to COMP6685, appropriate for a master level
The University is committed to ensuring that core reading materials are in accessible electronic format in line with the Kent Inclusive Practices.
The most up to date reading list for each module can be found on the university's reading list pages.
Bengio, Yoshua, Ian Goodfellow, and Aaron Courville. Deep learning. MIT press, 2017.
Kelleher, John D. Deep learning. MIT press, 2019.
https://www.deeplearningbook.org/
Sejnowski, Terrence J. The deep learning revolution. MIT press, 2018.
The intended subject specific learning outcomes.
On successfully completing the module students will be able to:
1 Comprehend the benefits of data re-representation in deep neural networks, and their ensuing modelling capacity.
2 Demonstrate an understanding of the algorithms that are required to train deep neural networks.
3 Demonstrate an awareness of computational and practical challenges existing in deep learning.
4 Demonstrate a systematic understanding of the key parameters in a neural network's architecture.
5 Competently use deep learning software to solve practical problems.
6 Understand the objectives of explainability and interpretability of neural networks.
7 Explain the differences between the major deep learning architectures.
8 Critically evaluate the strengths and weaknesses of the state-of-the-art deep learning models and algorithms.
The intended generic learning outcomes.
On successfully completing the module students will be able to:
1 Demonstrate critical thinking and problem-solving skills.
2 Communicate with other professionals using appropriate technical vocabulary.
3 Construct reasoned arguments about pros and cons of algorithms and their implementations.
4 Demonstrate originality in tackling and dealing with challenges related to the process of generalisation from data in a broader context of scientific enquiry.
University of Kent makes every effort to ensure that module information is accurate for the relevant academic session and to provide educational services as described. However, courses, services and other matters may be subject to change. Please read our full disclaimer.
