1 CE889 – Artificial Neural Networks Assignment Autumn 2020 1. Objectives To translate the theoretical knowledge gained throughout the course into practise by designing and implementing a neural network capable of solving a dynamic environment problem. To work in your groups to provide a Deep Neural Network entry to a Kaggle Competition. 2. Deadline and Submission Requirements The deadline for the assignment is as specified by the School Assignment deadline. Demonstration of the project: There will be two demonstrations, one to demonstrate your individual Neural Network in a one-on-one scenario, and the other to demonstrate your groups Deep Neural Network. These demonstrations will take place at the same time during your lab session scheduled for week 11 and will last ~30 minutes in total. Therefore, you must be organised as a team and ensure that everyone is available to attend for the entire length of the demonstrations. During the demonstrations students will be required to present their code in its finished state (which you will send to us prior to the demonstration), explain its main features, answer questions relating to their work and the project, and discuss the performance of the neural networks. 3. Assessment Criteria Performance (100%): 50% - The performance of your individual Neural Network. 50% - The performance of the groups Deep Neural Network in the Kaggle competition. 4. Important Information You must work independently on your individual neural network and this must be constructed using Python and use no additional libraries. 2 You may use any programming language and libraries your group agree upon for the team project. Any late submission will receive a Zero mark unless extenuating circumstances apply. Your work must be sent to a GLA prior to your demonstration so that it can be tested on the demonstration day. The demonstrations will take place over zoom. 5. The Task Individual task: You are presented with a lander game which will randomly generate a landing zone and unsafe terrain. You will be required to safely land the ship without touching the unsafe terrain. Data will be generated following your completion of this task; you must use this data to successfully train your own neural network with the ability to complete the same task. 1) Design and implement a Feed-Forward Backpropagation neural network where: i. Inputs: 8 Lander sensors, Landers current angle, X position to target, Y position to target ii. Outputs: Lander thruster, Lander turning 2) Train your neural network in your Python implementation (offline training ~100 epochs) with the data you have collected during your attempt at the lander game. 3) Test your neural network by running the most recent weights (just feed- forward) to see how the lander performs and compute the Root Mean Squared Error. Group Task: You are asked to work in your allocated groups of three to provide a Deep Neural Networks entry to the Rossmann Stores sales Kaggle Competition (https://www.kaggle.com/c/rossmann-store-sales). The competition aims to predict the Rossmann 1,115 stores sales across Germany. You are asked to do the following: 1) Choose one of the available Deep Neural Network tools and develop a Deep Neural Network that will be trained over the supplied data to provide an entry for the Rossmann Stores Sales competition. 2) Provide your results over the supplied testing data.
欢迎咨询51作业君