COMP2712 : Classifying Images Using MLP and SVM Algorithms - IT Computer Science Assignment Help

Assignment Task:

Task:

Classifying Images Using MLP and SVM Algorithms (25%) This assignment is to explore the concepts covered in the topic so far with respect to training and evaluating classifiers such as the Multi-layer Perceptron (MLP) and Support Vector Machines (SVM). The tasks and the associated weightings towards the final marks are outlined below.

1. Load the CIFAR-10 dataset

2. Pre-processing a. Normalisation 5% b. PCA exploration 15%

3. Classification Exploration a. ANN/MLP 15% b. SVM 15%

4. Evaluation a. 10 x 10 CV 15% 5. Discussion 35% Starter Google Colab Notebook There is a starter Google Colab Notebook that should be read in conjunction with this document and can be used as a starter for this assignment. It contains examples for reading and manipulating the dataset and training a standard MLP. The start notebook can be found here: https://colab.research.google.com/drive/1G1TNgylOhanqFD3uvruD2BLdNbTFngv3?usp=sharing You are to use the scikit-learn MLP and SVM implementations.

1. Load dataset The dataset for this assignment is the CIFAR-10 dataset that can be found here: https://www.cs.toronto.edu/~kriz/cifar.html The CIFAR-10 and CIFAR-100 are well studied, yet challenging image recognition dataset. The CIFAR10 has up to 10 classes to classify and contains 60,000 32x32 images. You should read the description of the dataset and download the dataset for Python, that is CIFAR-10 python version: https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz Once downloaded you need to then extract and upload the cifar-10-batches-py directory your Google Drive so that you can access it from within your Google Colab. The supplied notebook has an example of how to then read the data and to classify using an MLP on the full 10 classes, as well as classifying just 2 classes. However, the performance on the 10-class problem using the MLP on unseen test data is very poor at around 30% accuracy with most many images being classified as an airplane! Your task will be to improve upon this base level.

2. Pre-processing (20%) As has been demonstrated in the labs, pre-processing of the data can make for more efficient and effective optimisation. Your objective for this task is to apply normalisation (5%) and then explore COMP2712_8715 Assignment01 Page 2 of 2 Principle Components Analysis (PCA, 15%) based analysis to investigate if we can apply feature reduction to this classification problem. With the 32x32 image size and 3 colour channels the number of raw features is 3,072. Can the classifier do better with less features? How many components would be useful? Is there any pattern in features contributing to the components?

3. Classification Exploration (30%) For this assignment you need to explore the two types of classifiers of Multi-layer perceptron (MLP, 15%) and Support Vector Machines (SVM, 15%). The MLP should have at least one hidden layer with fully connected layers. You should investigate which parameter settings for each classifier result in the best classification. You should consider using a nested cross-fold validation to discover these settings.

4. Evaluation (15%) A key consideration for any machine learning optimisation expedition is correctly evaluating and comparing your models. You need to ensure that the results reported are on unseen test data to better measure the generalisation performance of the models. The most comprehensive approach to doing this is to perform a 10x10 cross fold validation and report the mean performance and its standard deviation or standard error. You should consider testing for a statistical significance.

5. Discussion (35%) Your overall mark will be based both on the implementation in the Google Colab Notebook and on the presentation and discussion of your results.

The above IT Assignment has been solved by our  IT Assignment  Experts at My Uni Paper. Our Assignment Writing Experts are efficient to provide a fresh solution to this question. We are serving more than 10000+ Students in Australia, UK & US by helping them to score HD in their academics. Our experts are well trained to follow all marking rubrics & referencing style.

Be it a used or new solution, the quality of the work submitted by our assignment experts remains unhampered. You may continue to expect the same or even better quality with the used and new assignment solution files respectively. There’s one thing to be noticed that you could choose one between the two and acquire considered worthy of the highest distinction.