The final project will be a team effort with 3-4 team members. We recommend you stay with your current project group, but you are allowed to change teams if you have a specific project idea / team members in mind. For your project, you need to use Machine Learning on an application specific area of your choice. Given the relatively short time to complete it, developing an idea for the project that is complex enough to be interesting and also within scope is crucial.
While no project is expected to do well on the entire Machine Learning pipeline, your team needs to choose what aspects you will prioritize on. You can start by looking at this list of parts of the Machine Learning Pipeline to address it. When developing your project, you should think about what your most important goal is and which of the following you will spend the most effort on.
Pick both aMaincomponent from the list below and aSecondarycomponent:
Data Collection / Curation- Examples include collecting data through cameras, sensors in phones (infrared, depth sensors, lidar), scraping data from web, collecting data in the wild (in workshop settings, in farms, in forests, in households, in parks, cities, etc), reforming and meaningful subsets of large public repositories, creating data with a labeling project.
Feature Engineering- Transforming raw data using manual, algorithmic, or machine Learning to enable better results over a baseline. Doing various steps for transforming raw data – signal conditioning, transforming to frequency domain, wavelet transforms, connected components gaussian blur, edge detection (if using cameras), DCF (discriminative correlation filters). Extracting domain-specific features, autocorrelation, octaves, band ratios, that are specific to datasets. Examples also include applying word embedding models, multiple transformations from larger models.
Developing a Model- Applying a framework that creates a model that can perform Connected CompRegression, Classification, or Clustering (PCA, K-means, etc). You may use data from existing resources (e.g. Kaggle).
Evaluation- Examples include rigorous comparisons of existing techniques under different conditions, testing multiple kinds of Machine Learning approaches, and developing synthetic data for testing how robust existing Models are. In-the-wild classification or “live” or real-time classification of developed models, testing for accuracies across different contexts (e.g., model developed in one room, being deployed in another room in building), testing across multiple users, leaving one out user study, user dependent vs independent model performance.
Applications of Existing Models- Using open-source models to apply to a novel context. For example, creating a demo that applies image generation, using sound detection algorithms in a specific classification setting, etc. Here, if there is a compelling demo, the report and evaluation can be scoped down.
You need to declare which of the above might constitute major efforts of the project so that your instructors can evaluate your results appropriately. For this class, you should not attempt to deeply perform all of the above, but we do expect your team to pick a secondary effort. While it is difficult to quantify, roughly 50% of your effort might address the main component and 25% of your effort may be spent on the secondary component and 25% on the rest.
Submission Structure:
Develop a Project Proposal Abstract:
- The abstract is a 150-200 word description of the proposed work written as if you have done it already. You may check prior example papers (e.g. from Assignment 2) for examples. The goal of the abstract is to introduce your topic, the work you plan to do, and the "payoff" or expected benefit.
- Worth 5% of total final grade
Get Feedback from Instructors:
- The instructors will read your abstract and offer suggestions to aid in planning and scoping your project. You may be asked to add additional work, narrow the project to shorter goals, or use this time to begin a proposal that will be used in your Project Proposal.
Project Proposal:
- The project proposal is a 1-2 page expansion of the abstract that includes a weekly plan of what your team will do. This should follow the template (to be provided) that describes the project timeline and its goals.
- Worth 5% of total final grade
Project Development:
- Time to work on your project and presentation.
Project Presentations:
- Short video presentation and report that includes your modified proposal and final results. Depending on time, live presentations with a Question and Answers section may be included.