Soft Computing Class Semester 2 2023 Assignment 1: Neural Networks (Worth 20 marks towards your assignment mark) Due Dates: Assignment 1 is due (submission on Canvas) on: Sunday 1^st October at 11:00pm. You need to get started early. For this assignment you will do up to 4 tasks using Python and its frameworks (PyTorch or TensorFlow). A one-paragraph written report/form is required for each task. Grades will be reduced for excessively long written submissions. You will need fill required tables and write one paragraph for each task. Upload codes and report file in zip format at submission portal. This needs to be done individually. You may discuss what you are doing with others in the class, BUT YOU MUST DO THE WORK YOURSELF. Tasks 2, 3, and 4 are slightly different for Undergraduate and Postgraduate. Please read information for each task to see the difference. For data for assignemet1: please download the zip file Data_Assignment1.zip and extract it. For each task there is a folder which you will use data for each task from there. Task1 Train a basic MLP Using the Dermatology dataset train a neural network using the 70-30. Split into test and train. Plot test and training loss function for epoch. Fill the following table and write a paragraph for it. Filling out the following form is suitable as a one-page submission for this task.

Details of MLP	Hidden Nodes: Epochs: Hyperparameters: Testing Accuracy: Training Accuracy:
Best estimate of test accuracy for a generalised solution.

Task2 Train a basic CNN and MLP

2. Using dataset for prediction of music genre.
4. Using diabetes dataset for binary classification.

UG Students: Train a CNN neural network on one dataset (number 1 or number 2) of the data sets provided. (use just one of them). Then train a MLP on the selected dataset which you trained CNN on that and this time train on MLP. PG Students: Train a CNN neural on all two (number 1 and number 2) of the data sets provided. Train a MLP on the both data sets. You must determine your best estimate of the accuracy of the resultant classifier for a generalised (i.e. not overtrained or under trained) solution. You should use the 70% train, 30% test data split methodology. Filling out the following form is suitable as a one-page submission for this task.

Dataset name
Details of CNN	Epochs: Hyperparameters: Testing Accuracy: Training Accuracy:
Details on MLP	Epochs: Hidden Nodes: Hyperparameters: Testing Accuracy: Training Accuracy:
Best estimate of test accuracy for a generalised solution for CNN.
Best estimate of test accuracy for a generalised solution for MLP.

Second table is just for postgraduate students. Filling out the following form is suitable as a one-page submission for this task.

Dataset name
Details of CNN	Epochs: Hyperparameters: Testing Accuracy: Training Accuracy:
Details on MLP	Epochs: Hidden Nodes: Hyperparameters: Testing Accuracy: Training Accuracy:
Best estimate of test accuracy for a generalised solution for CNN.
Best estimate of test accuracy for a generalised solution for MLP.

Task3 Build a pretrained model for Chest-Xray dataset For UG students: use one pretrained model to train and validate the model by using ChestXray image data set. Use train set for training and val set for validation. For PG students: use two pretrained model to train and validate the model by using ChestXray image data set. Use train set for training and val set for validation. Compare the results of the both pretrained models. Filling out the following form is suitable as a one-page submission for this task.

Pertained name
Details of model	Epochs: Hyperparameters: Testing Accuracy: Training Accuracy:
Best estimate of test accuracy for a generalised solution for pretrained model.

Second Table is for PG students. Filling out the following form is suitable as a one-page submission for this task.

Pertained name 2
Details of model	Epochs: Hyperparameters: Testing Accuracy: Training Accuracy:
Best estimate of test accuracy for a generalised solution for pretrained model.

Task4 Build a LSTM to time series analysis. Dataset number 1. Miles_Traveled.csv Dataset number 2: Alcohol_Sales.csv For UG students: use one dataset (number 1 or number 2) to train and validate (70 to 30) the LSTM model. For PG students: use two dataset (number 1 and number 2) to train and validate (70 to 30) the LSTM model. Filling out the following form is suitable as a one-page submission for this task.

Details of LSTM model	Epochs: Hyperparameters: Testing RMSE: Training RMSE:
Best estimate of test RMSE for a generalised solution for pretrained model.

For PG students only: Filling out the following form is suitable as a one-page submission for this task.

Details of LSTM model	Epochs: Hyperparameters: Testing RMSE: Training RMSE:
Best estimate of test RMSE for a generalised solution for pretrained model.