Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

Question 1. Convolutional Neural Networks (CNNs)

You work for the digital start-up company “New Healthy U” developing a food nutrition app, which will recognise different types of food in photographs. You are given an assignment by the CTO to analyse several CNN-based machine learning approaches for accomplishing this task and to produce a report containing your findings and recommendations. This report will influence the decision of the CTO on the approach to take in the app development. You will test several CNN models and classification methods using a publicly available dataset, which is a standard approach for testing CNNs.

Download the whole data set “Food-11 image dataset” if you have at least 1GB free on your hard drive from https://www.kaggle.com/datasets/trolukovich/food11-image-dataset. Alternatively, download a subset of 30 training, 5 validation and 5 test images per class.

Perform the following tasks on the downloaded dataset in Matlab:

i) Study the dataset and explain the data and its main parameters in the report.

ii) Use AlexNet as a feature extractor for all images in your training dataset containing 30 images per class. Use the extracted features to train a simple NN or another type of classifier (for example, SVM, KNN, neural network etc.) to classify the food into one of eleven classes. Use 30 training images for training, 5 validation images for validation and 5 test images for testing.

iii) Apply transfer learning to AlexNet so that it classifies the food into one of the eleven classes. Use the training, validation and test images in the same way as in experiment ii).

iv) Repeat experiments ii) and iii) using another pre-trained CNN, for example, GoogLeNet.

Adjust the size of the images to match the input size of the above networks. Refer to the lab examples on how this can be done. Choose and report appropriate parameters for classifiers, CNN training process, etc. Justify your choices. Report training, validation and test accuracy in all experiments. Research other ways of analysing the results, apply them to your results and add your analysis to the report. Discuss the results. Which experiments produced the best test accuracy and why? Make appropriate conclusions. Provide a variety of graphs to illustrate the experiments and their results. The marks will be based on the quality of the report with special attention given to the presentation and analysis of the results and the discussion. Include in your report screenshots of Matlab and Deep Network Designer - produced graphs and other output demonstrating your work throughout the whole processes of data preparation, network design, options setting, and network training and testing. [25]

Question 2. LSTMs

The “New Healthy U” company is very ambitious in several areas of digital health monitoring. Another project in its portfolio concerns monitoring physical activity and calculating calorie expenditure of people performing daily tasks of living. You are given an assignment by the CTO to investigate whether an LSTM network would be suitable for accomplishing this task and to produce a report containing your findings and recommendations. This report will influence the decision of the CTO on the approach to take in the app development. You will test LSTM performance when applied to recognising simple human activities from the accelerometer data obtained using a smartphone. You will use a publicly available dataset, which is a standard approach for testing machine learning algorithms.

Download the “Smartphone-Based Recognition of Human Activities and Postural Transitions Data Set” from http://archive.ics.uci.edu/dataset/341/smartphone+based+recognition+of+human+activities+and+postural+transitions

Perform the following tasks on the downloaded dataset in Matlab:

i) Study the dataset and explain the data and its main parameters in the report.

ii) Using Matlab and one of the examples from the LSTM lab, design and train an LSTM network on the training subset of the downloaded data. Report on the design of the LSTM network. Split the test dataset into two equal parts and use the first part for validation and the second for testing.

iii) Choose and report appropriate parameters for LSTM training and justify your choice. Report the achieved training, validation and test accuracy. Research other ways of analysing the results, apply them to your results and add your analysis to the report. Discuss the results. Make appropriate conclusions. Provide a variety of graphs to illustrate the experiment and the results. The marks will be based on the quality of the report with special attention given to the presentation and analysis of the results and discussion. Include in your report screenshots of Matlab - produced graphs and other output demonstrating your work throughout the whole processes of data preparation, network design, options setting, and network training and testing. [25]