ÐÂÏã¸ÛÁùºÏ²Ê¿ª½±½á¹û

Outline:

The module aims to provide an introduction to the use of machine learning in the context of physics data handling and analysis. Topics: Fundamental concepts in machine learning and modern statistics such as bias-variance, overfitting, regularization and generalization. Advanced topics in supervised and unsupervised learning: ensemble models, deep learning, neural networks, clustering and data visualization.

Aims:

After taking this module, students should:

• understand the key concepts of machine learning;
• know how to create clear, reliable and maintainable code;
• be able to utilize machine learning and artificial intelligence techniques to analyse datasets such as those that arise in level-6 laboratories and level-6 and level-7 projects.

Intended Learning Outcomes:

At the end of this module the student should:

• Understand the origin and use of neural networks.
• Understand supervised neural network training and the back propagation algorithm.
• Understand the role of machine learning in image recognition and data compression.
• Be able to design and train networks using the Tensorflow + Keras.
• Understand and be able to use convolutional neural networks and auto encoders.
• Be able to utilise principal component analysis and other dimensional reduction techniques and unsupervised learning methods.
• Understand the problems of modelling sequential data and use recurrent networks to address these problems.
• Understand the basic concepts of reinforcement learning, including Markov decision processes and value and policy based learning approaches.
• Understand how ideas from statistical thermodynamics lead to Boltzmann machines and energy based models.
• Understand how to use and build generative models.
• Understand some of the ways machine learning is used in physics research.

Teaching and Learning Methodology:

The course is taught via lecture and interactive practical sessions. All of the session time in the course will be devoted to hands-on practice on the computers.

In addition to timetabled lecture hours, it is expected that students engage in self-study in order to master the material. This can take the form, for example, of practicing example questions and further reading in textbooks and online.

Indicative Topics:

Neural networks, gradient descent, back propagation, data compression, image classification, convolutional neural networks, auto encoders, supervised learning, unsupervised learning, reinforcement learning, dimensional reduction, principal component analysis, stochastic neighbour encoding, sequence modelling, energy based models, generative models, generative adversarial networks.