Hackathon in AI for Sustainability 2022
The other week, I took part in the Hackathon in AI for Sustainability 2022. While this was notable because it was my first hackathon, what was more important was that it was partially based on my research! For those who aren't aware, I'm currently doing a PhD at the University of Hull with the project title "Using Big Data and AI to Dynamically Predict Flood Risk". While part of it really hasn't gone according to plan (I do have a plan to fix it, I just need to find time to implement it), the second half of my project on social media has been coming together much more easily.
To this end, my supervisor asked me about a month ago whether I wanted to help organise a hackathon, so I took the plunge and said yes. The hackathon has 3 projects for attendees to choose from:
- Project 1: Hedge identification from earth observation data with interpretable computer vision algorithms
- Project 2: Monopile fatigue estimation from nonlinear waves using deep learning
- Project 3: Live sentiment tracking during floods from social media data (my project!)
When doing research, I've found that there are often many more avenues to explore than there is time to explore them. To this end, a hackathon is an ideal time to explore these avenues that I have not had the time to explore previously.
To prepare, I put together some dataset of tweets and associated images - some from the models I've actually trained, and others (such as one based on the hashtag
#StormFranklin) that I downloaded specially for the occasion. Alongside this, I also trained and prepared a model and some sample code for students to use as a starting point.
On the first day of the event, the leaders of the 3 projects presented the background and objectives of the 3 projects available for students to choose from, and then we headed to the lab to get started. While unfortunate technical issues were a problem for all 3 projects, we managed to find ways to work around them.
Over the next few days, the students participating in the hackathon tackled the 3 projects and explored different directions. At first, I wasn't really sure about what to do or how to help the students, but I soon started to figure out how I could assist students by explaining things, helping them with their problems, fetching and organising more data, and other such things.
While I can't speak for the other projects, the outputs of the hackathon for my project are fascinating insights into things I haven't had time to look into myself - and I anticipate that we'll be may be able to draw them together into something more formal.
Just some of the approaches taken in my project include:
- Automatically captioning images to extract additional information
- Using other sentiment classification models to compare performance
- Resolving and inferring geolocations of tweets and plotting them on a map, with the goal of increasing relevance of tweets
The outputs of the hackathon have been beyond my wildest dreams, so I'm hugely thankful to all who participated in my project as part of the hackathon!
While I don't have many fancy visuals to show right now, I'll definitely keep you updated with progress on drawing it all together in my PhD Update blog post series.