Computing

Machine Learning

Module code: 934G5
Level 7 (Masters)
15 credits in spring teaching
Teaching method: Laboratory, Lecture
Assessment modes: Coursework

In this module, you explore advanced techniques in machine learning.

You use a systematic treatment, based on the following three key ingredients:

  • tasks
  • models
  • features.

As part of the module, you are introduced to both regression and classification, and your studies emphasise concepts such as model performance, learnability and computational complexity.

You learn techniques including:

  • probabilistic and non-probabilistic classification and regression methods
  • reinforcement learning approaches including the non-linear variants using kernel methods.

You are also introduced to techniques for pre-processing the data (including PCA).

You will then need to be able to implement, develop and deploy these techniques to real-world problems.

In order to take this module, you need to have already taken the 'Mathematics & Computational Methods for Complex Systems' module (817G5), or have taken an equivalent mathematical module or have equivalent prior experience.

Pre-requisite

Mathematics & Computational Methods for Complex Systems (817G5) or equivalent mathematical module / prior experience

Module learning outcomes

  • Identify the strengths and weaknesses of state-of-the-art supervised, unsupervised, and reinforcement machine learning models including multi-layer perceptron, support vector machine, random forest, K-means, PCA, and Q-learning.
  • Critically analyse and implement several stochastic optimization methods ranging from stochastic gradient descent, stochastic variance reduction, to adaptive gradient methods for training machine learning models on big data.
  • Demonstrate knowledge of the fundamental principles of advanced machine learning models including probabilistic graphical models and statistical network models.
  • Apply developed classification/regression techniques with stochastic optimization to real-world problems, including extracting deep convolutional neural network features and incorporating prior knowledge.