Project
The final project is designed to give you experience implementing a machine learning model on a real-world dataset—without me telling you what method to use. We want you to explore any creative or unconventional methods you can think of! This is a completely open-ended problem, and the most important thing I want to see is effort. So let’s go over the project details:
Topics
There are three project topics covering a range of machine learning tasks:
- Identify Chinese characters (computer vision) — You will be given images containing a mix of handwritten numeric digits and Chinese characters. Your task is to identify and draw bounding boxes around the Chinese characters.
- Informative tweets (sentiment analysis) — Based on the WNUT-2020 competition, you are given a dataset of tweets related to the COVID-19 pandemic. Your job is to identify which tweets are informative.
- Virtual bidding in electric markets (trend prediction) — Develop a classification algorithm and trading strategy to identify profitable trading opportunities.
Dataset
The datasets are hosted on GitHub here.
Kaggle Competition
To make things more exciting, we are hosting a competition on Kaggle! You will submit your best runs to Kaggle (details are in the project description files).
The top individual or team for each topic will receive a prize:
- Individuals: Full letter grade bump
- Teams (two or more members): Half-letter grade bump
We hope this will encourage you to try new methods and explore the current state-of-the-art models in machine learning.
We will be verifying the winners and ensuring the submitted numbers come from models that follow the competition rules.
Important: Make sure to set a random seed so that your training is reproducible.
Teams
You may work in teams of up to three people and only need to complete one of the three project topics. There’s a variety to choose from, so pick the one that interests you most!
Deliverables
You must submit:
- Your code
- A two-page report describing what you did, what you tried, whether it worked, and what you would do differently next time.
Due Dates
There are two important deadlines:
- Kaggle competition closes: May 5th (Monday before Reading Day)
- Two-page report due: May 12th (Monday)
Winners!
The ECE364 project for the Spring 2025 semester has officially concluded! We were genuinely impressed by the creativity and enthusiasm students showed in experimenting with different model combinations and training strategies. Across all tasks, there were teams that surprised us with their innovative approaches and strong results. Here are a few personal takeaways from the project:
- Task II was the most popular, followed by Task III, and then Task I. I honestly expected the reverse—but the data speaks for itself.
- In Task II, several teams came very close to the best possible performance. It’s fascinating how often these competitions are decided by the slimmest of margins.
- In Task III, many teams approached the theoretical maximum performance! Looks like we’ll need to design more complex market environments next time.
- Interestingly, there was no correlation between team size and competition performance.
Anyway, here are the winners of each task competition and the teams that provided the best reports:
| Task | Competition Winner | Best Report Winner |
|---|---|---|
| Task I (Vision) | Junyi Chen, Shengyu Xu, Chaohua Yao | Jiwoong Jung, Pump Vanichjakvong, Haoming Mei |
| Task 2 (NLP) | Alayna Li, Ayush Barik, Brent Alonzo | Ding Jiang, Denghan Xiong, Yaxing Zhang |
| Task 3 (Markets) | Riley Douglas | Vikram Chakravarthi |