Artificial Intelligence & Machine Learning - Module CE6002

The module will cover the following areas

• The learning problem: feasibility of learning, error, and noise
• Theory of generalization: Effective number of hypotheses, VC bound, sample and model complexity, approximation-generalization trade-off, bias and variance
• Linear classification and regression, logistic regression, gradient descent, and feature space transformations
• Overfitting and regularisation
• Validation and model selection, data snooping
• Neural Networks: Perceptron’s, Multi-Layer Perceptron’s and the Back-Propagation training algorithm.
• Optimal Margin Classifiers and Support Vector Machines.
• Parametric vs. Non-Parametric classifiers.

On successful completion of this module students will be able to:

• Demonstrate an understanding of the theory of generalisation and its practical implications for machine learning algorithms, the concept of model complexity, in particular the VC bound and its practical interpretation.
• Be able to apply regularization in order to prevent overfitting.
• Demonstrate an understanding of and be able to apply non-linear transformations to feature spaces.
• Recognise and manage under- and overfitting.
• Apply methods for selecting one final machine learning model from among a collection of candidate machine learning models for a training dataset.
• Apply methods for model validation, the process where a trained model is evaluated with a testing data set.
• Apply a number of linear and non-linear and parametric and non-parametric machine learner training models e.g., linear regression, logistic regression, feed forward neural networks and Support Vector Machines.
• Differentiate and critique various techniques that could be used and be able to justify an appropriate classification technique for a given a classification problem.
• Demonstrate an awareness of and be able to implement appropriate protocols and practices to manage bias and data snooping when training a machine learner, for a given data set.
• Demonstrate an awareness of the impact of the availability of data, for a given data set used to train the machine learner, when assessing the machine learner's performance.