UNIT INFO

COMSM0045 - Applied Deep Learning

ADL Banner

Unit Information

Welcome to COMSM0045. The unit introduces the students to deep architectures for learning linear and non-linear transformations of big data towards tasks such as classification and regression. The unit paves the path from understanding the fundamentals of convolutional and recurrent neural networks through to training and optimisation as well as evaluation of learnt outcomes. The unit's approach is hands-on, focusing on the 'how-to' while covering the basic theoretical foundations. For further general information, see the syllabus for the unit.

UPDATE - 15/09/2023 Welcome new students for the 23/24 year!

Staff

Michael Wray (MW)Unit Director
Tilo Burghardt (TB)

Teaching Assistants

Adriano Fragomeni (AF), Ahmad Dar Khalil (AK), Dan Whettam (DW), Jacob Chalk (JC), Kevin Flanagan (KF), Sam Pollard (SP), Saptarshi Sinha (SS), Siddhant Bansal (SB), Zhifan Zhu (ZZ)


Unit Materials

Wks Tues 09:00-11:00 Friday 09:00-12:00 Labs
1 26/09/2023 - 9am - Queens BLDG 1.69
Wk1 - LECTURE 1
INTRODUCTION TO THE UNIT
intro PDF

BASICS OF ARTIFICIAL NEURAL NETWORKS
(Queens Building 1.69, in-person)
(Introduction, Neural Networks, Perceptron, Cost Functions, Gradient Descent, Delta Rule, Deep Networks)
PDF Slides
29/09/2023 - 9am - Queens BLDG Bill Brown Suite
Wk1 - LECTURE 2
TOWARDS TRAINING DEEP FORWARD NETWORKS
(Queens Building F.101c (Bill Brown Suite), in-person + lecture recap)
(Network Representation, Computational Graphs, Reverse Auto-Differentiation)
PDF Slides, Extra Recap Recording Lecture 2 Refresher (first part of video)
GETTING STARTED:

Register Individually on BlueCrystal4
(details see below)

RECAP WORKSHEETS:
-Convolutions (Homework)
-Lab0 - Python (Homework)
2 03/10/2023 - 9am - Queens BLDG 1.69
Wk2- LECTURE 3
BACKPROPAGATION ALGORITHM
(Queens Building 1.69, in-person + recorded lecture)
(The Backpropagation Algorithm in Full Detail, Activation Functions)
PDF Slides
Extra Recap Recording (second part of video)
Wk2 - LECTURE 4

OPTIMISATION TECHNIQUES
(Queens Building 1.69, in-person + recorded lecture)
(Stochastic Gradient Descent, Nesterov Momentum, RMSProp, Newton's Method, AdaGrad, Adam, Saddle Points)
PDF Slides
Extra Recap Recording
06/10/2023,9am,(Queens BLDG Bill Brown Suite) - PRACTICAL 1 (Slides),
Your first fully connected layer
gradient descent
stochastic gradient descent
06/10/2023, (Queens BLDG Bill Brown Suite) - 3hrs
-BC4 Setup
Lab 1 - Training your first Deep Neural Network
3 10/10/2023 - 9am - Queens BLDG 1.69
Wk3 - LECTURE 5
(Queens Building 1.69, in-person + recorded lecture)
COST FUNCTIONS, REGULARISATION AND DEPTH
(SoftMax, Cross Entropy, L1 and L2 Regularisation, DropOut, DropConnect, Depth Considerations)
PDF Slides
Extra Recap Recording
13/10/2023,9am,(Queens BLDG Bill Brown Suite) - PRACTICAL 2 (Slides), (Video).
Your first convolutional connected layer
13/10/2023, (Queens BLDG Bill Brown Suite) - 3hr
Lab 2 - Your First Convolutional Connected Network
Wk3 - LECTURE 6
(Queens Building 1.69, in-person + recorded lecture)
CONVOLUTIONAL NEURAL NETWORKS
(Video Pt. 1, Video Pt. 2, Slides)
(sharing parameters, conv layers, pooling, CNN architectures)
4 - 20/10/2023,9am,(Queens BLDG Bill Brown Suite) - PRACTICAL 3 (Slides, Video Recording)
Error rate monitoring (training/validation/testing)
Batch-based training
Learning rate
Weight Freezing
Batch normalisation
Parameter intialisation
20/10/2023, (Queens BLDG Bill Brown Suite) - 3hr

Lab 3 - Hyperparameters
5 - 27/10/2023, 9am, (Queens BLDG Bill Brown Suite)
- PRACTICAL 4
Data Augmentation
Debugging strategies
Dropout (Slides), Video Recording

27/10/2023, (Queens BLDG Bill Brown Suite) - 3hr

Lab
Lab 4 - Data Augmentation
6 READING WEEK (no activities)
7 07/11/2023 - 9am - Queens BLDG 1.69
Wk8 - LECTURE 7
(Queens Building 1.69, in-person + Recorded)
RECURRENT and RELATIONAL NEURAL NETWORKS
(RNN, bi-directional RNNs, encoder-decoder, Transformers)
(Slides, Video)
Wk 8 - LECTURE 8
(Queens Building 1.69, in-person + Recorded)
GENERATIVE MODELS
(GANs, Diffusion Models)
(Slides, Video pt. 1, Video pt. 2)
10/11/2023, 9am, (Queens BLDG Bill Brown Suite) - PRACTICAL 5
Transformers
(Slides, Video Recording)

10/11/2023, (Queens BLDG Bill Brown Suite) - 3hr
Lab 5
Transformers
8 - 17/11/2023, 9am, (Queens BLDG Bill Brown Suite)
Continuation Lab + First CW Lab
17/11/2023, (Queens BLDG Bill Brown Suite) - 3hr

9 - 24/11/2023, 10am [2 hours], (Queens BLDG Bill Brown Suite) - CW Support Session -
10 - 01/12/2023, 10am [2 hours], (Queens BLDG Bill Brown Suite) - CW Support Session -
11 CW Deadline Thurs 07 Dec 13:00 (Blackboard Submission)
12 - 15/12/2023, Queens BLDG Bill Brown Suite - Exam Support Session-

Assessment Details


Assessment Details - Coursework

Coursework specs can be found here.

Please fill in the form noting your chosen group (even if you're working solo) here

The data and provided code can be found here or on BC4 (see notes in coursework specs).

Assessment Details - Exam

2 hours exam in January (in person).

We have now released the exam preparation materials on the unit webpage, you can also find these in the folder here.

This includes a list of exam topics, 3 past papers (1 with answers) and example questions of new content this year.

Please note that you cannot take notes into the exam, but calculators are permitted.


Github

All technical resources will be posted on the COMSM0045 ADL Github organisation. If you find any issues, please kindly raise an issue in the respective repository.


Textbook

Recommended Reading:
Goodfellow et al (2016). Deep Learning. MIT Press


Blue Crystal 4 Registration [only applicable for Bristol undergraduate students with corresponding email]

All students must apply online to register an account on BC4 for this unit. This also applies to students who already have accounts on BC4 for other units (e.g. HPC), in this case you must register again using the instructions below.

  1. Click on: https://www.acrc.bris.ac.uk/login-area/apply.cgi
  2. Enter your personal details
  3. Choose: "Join an existing project"
  4. Enter project code: COMS030144
  5. Keep Preferred log-in shell as bash
  6. In the comments box please enter the following:
  7. "I am on the taught course Applied Deep Learning (COMSM0045), unit director Michael Wray, and will need access to BC4"

Note that it takes up to 48 hours to enable your account on BC4.