CSC8111 : Machine Learning

Semester 1 Credit Value: 10
ECTS Credits: 5.0


Machine Learning is concerned with the design of algorithms for recognising patterns in data. The field of pattern recognition represents the basis for a wide range of applications for automatic data analysis, such as computer vision, automatic speech recognition, or activity recognition – all based on sensor based observations of humans in their environment. The growth of “big data” means that such analysis techniques are now widely used for mining information from large amounts of data as they are collected in contemporary computing infrastructures, including clouds.
Conceptually, Pattern Recognition aims for the detection of instances of relevant classes that are typically associated with reappearing patterns in data streams. Examples of which are the automatic detection of faces in video streams, automatic transcription of spoken language, analysis of human movements, trend prediction in stock market data, intrusion detection in computer systems, or the analysis of social networks. The task is to find, model (or "learn") and classify those patterns, and to distinguish relevant from irrelevant events.
Machine Learning techniques represent the algorithmic foundation for such tasks, and involve both statistical modelling techniques and probabilistic reasoning approaches.
This module aims to provide a foundation in the field of Pattern Recognition and an expertise in Machine Learning techniques as a toolkit for automatically analysing (large amounts of) data – be it static data, such as images, or dynamic data, such as time series and sensor data.

Outline Of Syllabus

Introduction of Machine Learning: Supervised Learning, Unsupervised Learning, Data Representation, Overfitting
- Basics: Bayesian Theorem, Gaussian Distribution, Gaussian Mixture Models, Maximum Likelihood Estimation, Regularisation, Gradient Descent.
- Traditional Supervised Learning Models: Linear Regression, Logistic Regression (LR), Naive Bayes Classifier (NB), Decision Tree (DT), Random Forest (RF), Support Vector Machines(SVM), kernel SVM, K-nearest Neighbours Classifier (KNN).
- Traditional Clustering Methods: K-Means, Expectation Maximisation (EM)
- Deep Learning: Multilayer Perceptron(MLP), Convolutional Neural Network (CNN), Recurrent Neural Network(RNN), Long Short Term Memory (LSTM), ConvLSTM. Autoencoder (AE)
- Feature Extraction Methods: Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA)

Teaching Methods

Teaching Activities
Category Activity Number Length Student Hours Comment
Scheduled Learning And Teaching ActivitiesLecture221:0022:00Lectures
Guided Independent StudyAssessment preparation and completion161:0016:00Background Reading
Guided Independent StudyAssessment preparation and completion260:3013:00Revision for end of Semester exam and exam duration
Guided Independent StudyAssessment preparation and completion221:0022:00Lecture follow up
Scheduled Learning And Teaching ActivitiesPractical121:0012:00Practicals
Guided Independent StudyProject work151:0015:00Coursework
Teaching Rationale And Relationship

Lectures provide the algorithmic foundations of statistical pattern recognition and the practical work seeks to build on these foundations.
Each group presenting their distinct work to the rest of the class helps discern the capabilities of different modelling and analysis approaches.

Assessment Methods

The format of resits will be determined by the Board of Examiners

Description Length Semester When Set Percentage Comment
Written Examination901A80N/A
Other Assessment
Description Semester When Set Percentage Comment
Report1M20Individual work. 1000 words max.
Assessment Rationale And Relationship

The written examination will primarily focus on assessing a student's ability to analyse and use the structure and algorithmic foundations of statistical pattern recognition and machine learning systems.

Coursework assessment is through one individual deliverable, emphasising both the conceptual and applied nature of the module. Students will work on a single, but distinct, practical recognition task where they will set up and evaluate a recognition system that fulfils certain specified criteria. The deliverable is designed to allow students to demonstrate the conceptual underpinnings of the task being carried out and to provide a reflection on what they have learnt from it.

Study abroad students may request to take their exam before the semester 1 exam period, in which case the length of the exam may differ from that shown in the MOF.

N.B. This module has both “Exam Assessment” and “Other Assessment” (e.g. coursework). If the total mark for either assessment falls below 40%, the maximum mark returned for the module will normally be 40%.

Reading Lists