Machine Learning in Science MSc

This module will provide an introduction to the main concepts and methods of machine learning. It introduces the basics of supervised, unsupervised and reinforcement learning as applied to regression, classification, density estimation, data generation, clustering and optimal control. It will be taught via two sessions per week through a combination of fundamental concepts and hands-on applications.

This module will cover more advanced topics following from Machine Learning in Science Part 1, specifically the concepts and methods of modern deep learning. Topics include deep neural networks, CNNs, RNNs, GANs, RBMs and deep RBMs, autoencoders, transfer learning, reinforcement learning and Markov decision processes, cleaning data and handling large data sets. The main project for the module is the self-driving PiCar, as seen in this video.

You will carry out a substantial investigation in the form of a research project on the application of the machine learning techniques learned as part of the course to a scientific problem.

The study will be largely self-directed, with oversight and input provided by a supervisor from the School of Physics and Astronomy, School of Computer Science or School of Mathematical Sciences. The topic will be chosen from a list of potential projects provided by the schools in the Faculty of Science. The topic could be based on a theoretical and/or computational investigation, a review of research literature, and/or a combination of the two.

This module will begin by introducing a number of approaches to handling very large datasets, eg databases, indexing, chunking, parallelism, and map-reduce. You will look at some widely-used software tools which implement these methods.You will be introduced to the concept of cloud computing and give examples of the facilities provided by popular vendors. The module will also look at how big data algorithms can be implemented using cloud-based hardware, and finish with you deploying your own big data solutions in the cloud.

In this module, you will be given a basic introduction to the analysis and design of intelligent agents, software systems which perceive their environment and act in that environment in pursuit of their goals. You will cover topics including task environments, reactive, deliberative and hybrid architectures for individual agents, and architectures and coordination mechanisms for multi-agent systems.

You will examine current techniques for the extraction of useful information about a physical situation from individual and sets of images. You will learn a range of methods and applications, with particular emphasis being placed on the detection and identification of objects, image segmentation, pose estimation, recovery of three-dimensional shape and analysis of motion. These problems will be approached with both traditional and modern computer vision approaches, including deep learning.

This module looks broadly into professional ethics within the scope of the computing discipline. It covers a range of professional, ethical, social and legal issues in order to study the impact that computer systems have in society and the implications of this from the perspective of the computing profession. In particular, the module covers topics such as introduction to ethics, critical thinking, professionalism, privacy, intellectual and intangible property, cyber-behaviour, safety, reliability and accountability, all within the context of computer systems development.

The paradigm of Quantum Information Science (QIS) is that quantum devices, made of systems such as atoms and photons, can out-perform the present-day technology in key applications ranging from computing power and communication security to precision measurements. Quantum information processing and the measurement and control of individual quantum systems are central topics in QIS, lying at the intersection of quantum mechanics with 'classical' disciplines such as information theory, probability, and statistics, computer science and control engineering.

The aim of this module is to provide an introduction to QIS, emphasising the differences and similarities between the classical and the quantum theories. After a short review of the necessary probabilistic notions, the first part introduces the operational framework of quantum theory involving the fundamental concepts of states, measurements, quantum channels, instruments. This includes some of the influential results in the field such as entanglement and quantum teleportation, Bell's theorem and the quantum no-cloning theorem. The second part covers at least two topics from quantum Markovian evolutions, quantum statistics, continuous variable systems.

This module will teach you how to think about artificial neural networks and deep learning from the point of view of physic. It will apply concepts and methods of statistical mechanics to deep learning problems. It will show how a physical science perspective allows you to understand better the workings of neural networks to make them more efficient and expand their range of applications.

The aim of this module is to teach you how neural processes can be understood in computational terms and how they can be analysed using mathematical and computational methods.

Topics included:

• biophysical and reduced models of neurons
• models of networks (eg Hopfield networks, ring-attractors and rate networks)
• models of synaptic plasticity and memory
• perceptrons
• unsupervised learning
• neural coding
• visual system
• model fitting

Providing you with an introduction to machine learning, pattern recognition, and data mining techniques, this module will enable you to consider both systems which are able to develop their own rules from trial-and-error experience to solve problems, as well as systems that find patterns in data without any supervision. In the latter case, data mining techniques will make the generation of new knowledge possible, including very big data sets. This is now fashionably termed 'big data' science.

You'll cover a range of topics including:

• machine learning foundations
• pattern recognition foundations
• artificial neural networks
• deep learning
• applications of machine learning
• data mining techniques and evaluating hypotheses

This module is a topic at the interface between statistics and computer science, concerning models that can adapt to and make predictions based on data. This module builds on the principles of statistical inference and linear regression to introduce a variety of methods of clustering, dimension reduction, regression and classification.

Much of the focus is on the bias-variance trade-off and on methods to measure and compensate for overfitting. The learning approach is hands-on; you will be using R extensively in studying contemporary statistical machine learning methods, and in applying them to tackle challenging real-world applications.

In this module, the fundamental principles and techniques underlying modern statistical and data analysis will be introduced. You will gain experience in using a statistical package and interpreting its output.The course will cover a 'common core' consisting of:

• statistical concepts and methods
• linear models
• probability techniques
• Markov chains

In this module you will explore the fundamental principles and techniques underlying modern statistical modelling and data analysis. The course will cover a 'common core' consisting of statistical concepts and methods, linear models, probability techniques and Markov chains. You will gain experience of using a statistical package and interpreting its output.

In addition, you will study more advanced material concerned with the two main theories of statistical inference, namely classical (frequentist) inference and Bayesian inference. Topics such as sufficiency and best-unbiased estimators are explored in detail. There is special emphasis on the exponential family of distributions. Topics in Bayesian inference include basic ingredients (prior, likelihood and posterior), conjugacy, vague peior knowledge, marginal and predictive inference, decision theory, normal inverse gamma inference, and categorical data.

This module will give you a comprehensive overview of the principles of programming, including procedural logic, variables, flow control, input and output and the analysis and design of programs. Instruction will be provided in an object-oriented programming language.

This module will introduce the Python programming language and its associated ecosystem, with a focus on the elements most applicable to scientific computing. You will begin by covering the essentials of pure Python, as well as installation, environment maintenance and version control issues. You will then introduce the 'numpy' module for efficiently dealing with array data, followed by plotting with 'matplotlib', and the various scientific tools in 'scipy7'.

The module will continue with a tour of various modules useful for different aspects of scientific programming, including data handling and analysis, symbolic mathematics, Monte Carlo sampling, tools for machine learning, and producing graphical and online interfaces.

Finally, you will cover areas of good software development, such as testing and profiling, as well as approaches for dealing with very large amounts of data or speed critical applications.