Machine Learning and Data Mining (CETM26)   Assignment 2: Research Project   Submission ( 65% of Module Marks )   Aims   This assignment will pull from all aspects of the module, and explore both a research component alongside a practical component with Python

Machine Learning and Data Mining (CETM26)

Assignment 2: Research Project

Submission ( 65% of Module Marks )

Aims

This assignment will pull from all aspects of the module, and explore both a research component alongside a practical component with Python. This work will be a continuation of Neural Networks towards harder and more complex problems, and will include more in-depth techniques used when training networks.

Task - Hyper-parameter Optimisation

In assignment 1 you created a feedforward artificial neural network (ANN) to solve a binary classification task on tabular, numeric data. In this assignment we will expand upon these concepts, and solve a classification task on a harder problem involving image input (See Dataset section below). In this assignment, you will be expected to research techniques of your own accord, potentially beyond those that are taught within the module. This may involve state-of-the-art techniques in machine learning. You are expected to come up with a methodical and scientific approach to creating a model for solving multi-class classification of images, incorporating many of the techniques within the module such as different types of network, various architecture choices, hyper-parameter optimisation techniques, dimensionality reduction methods, etc. All of this will be written into a report outlining your project, comparing results, and evaluating your experiments. At the end of this report you will conclude your findings, and discuss considerations for these technologies towards the theme of ethical use of AI and how these systems may be both beneficial and/or disruptive; specifically in the context of the proposed solution for the CIFAR-10 dataset challenge and its potential applications.

Dataset

The dataset for this assignment is the CIFAR-10 dataset. The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images