# Final Project ### Making Team - 1~3 persons can make a team. For a larger team, a more challenging task is expected. - You can use the following link to find teammates: https://web.groupme.com/join_group/65606824/1iBYLIev ### Example project types You can do <!-- 1. **Reproducing results in the following paper**: [Mastering Atari with Discrete World Models](https://arxiv.org/abs/2010.02193) --> <!-- - Some related reference code is available: https://github.com/juliusfrost/dreamer-pytorch --> 1. **Reproducing results in papers.** You can try to reproduce the results of a research paper of your interest. This is applicable only when there is no code made available by the authors or someone else. 1. **Novel datasets and tasks.** You can develop a network to apply to a novel dataset or task. Because simply applying an existing model to a new dataset can be too boring, you should argue that why your task is a challenging task. 1. **Research.** You can define your own small research project, e.g., developing a new architecture or layer, analyzing some model, etc. 5. **Some combination of the above** <details><summary>Past project examples</summary> The following are examples from past projects - A Learned Manager-Worker Framework For Sampling-Based Motion Planning - Toward Robust Image Classification - Generative Video Transformer - CNN's for Breast Cancer Diagnosis - U-PC Net: Learning to Jointly Reconstruct and Segment Left Atrium in 3D from LGE-MRI Data - Spatial Attention for Medical Image Segmentation - Hidden Two-Stream Convolutional Networks for Action Recognition - Improvement over Ordered Neurons(ON-LSTM) - Learning to Explain Your Plan: Graph Neural Network-augmented Dyna - Graph attention auto encoders - Music generation with GAN and Transformer - Deep Learning for COVID-19 Diagnosis Based on CT Scans - Sign Language Interpretation - Learning grounded latent representation - semi-supervised named entity recognition </details> <!-- ### On reusing existing code - It is allowed to use existing implementations from e.g., GitHub. But, you have to specify the fact that which implementation you are using (by providing the link). Not claiming this means that you're implementing things from scratch. For each project, *TA will investigate the availability of relevant implementations*. - Simply using existing implementation without additional contribution is not a proper project topic. For example, if you use existing code on a different dataset, and found it works well, then it is not an interesting project (because there is not enough challenge). If it does not work well (so found a challenge), and you analyze the problem and make it better with your own idea, it is a good project. --> ## Proposal (==due by Apr 5, 11:59pm==) You are asked to submit the plan for your final project as a **1-page** report (references can be in the second page). An assignment will be created in Canvas through which you can upload your **pdf** file. **Even if you're doing a team project, each member of a team should submit a report individually (the same version as your teammates.)** The report should contain 1. Project title 1. Team member names and netids 1. Description - The problem statememt: **What** problem (challenge) you are going to solve - The approach plan: **How** you are going to solve - The evaluation plan: How your are going to **demonstrate** whether the goal (solving the problem) is achieved or not. Visualization and analysis are important factors. **Note that it is totally fine to change/update your plan after the proposal submission**. The goal of the proposal is to encourage you think about the project. **Formatting:** Use this latex template: [latex format file](https://www.overleaf.com/read/cxsbvxcwxjwq) ## Final Presentation (on May 5) Each team is allowed to present for **5 minutes**. The presentation should include - Introduction: what problem you tried to solve - Previous state of knolwedge: What have already been done in the past - Challenge: Why this problem is challenging - Main: what is the scientific contribution, how you solve the problem - Evaluation: how you evaluated your contribution - Conclusion Note that for reproducing projects, the contents of the presentation and report should be about the challenges that you encountered in reproducing the result and how you resolved it. Introducing the model in the original paper should not be the main focus because it's not what you make. You should also compare your result to the results reported in the oringinal paper. Upload your presentation slides in PDF format to Canvas by ==May 5, 3:00pm==. Each team has one proposal but every member should submit a copy of it individually. <!-- List your team members and project title [here](https://docs.google.com/spreadsheets/d/11zGtd2GD4OqIqZOlacPhgKrWDidkb-DL_JBSeKaP2NE/edit?usp=sharing) --> ## Final Report (due by May 8, 11:59pm) - Make the report look like a published academic paper as much as possible, including sections on Introduction, Related Works, Main Contribution, Experiment, Conclusion, References. The report should be in the following NeurIPS format and is limited up to 5 pages except for the references - Use latex: https://www.overleaf.com/latex/templates/neurips-2019/tprktwxmqmgk - No limit in the number of references - Can add Appendix for additional figures and results - If it is a team project, after the Conclusion section, add a section 'Contribution' where you specify what tasks each person in the team has contributed.