CodeML 2024

Fall of 2024, me and my friend from Marie-Victorin attended the CodeML 2024 conference in Montreal. This was our first time attending a AI oriented hackathon.

Multiple companies were present and they each presented a challenge to solve. We were given more or less 2 days to get good accuracy (F1-Score, R-Squared) on the models we were building. One of the challenges didn't require a model and was more of a ethical dillema to be solved around AI. Me and my teammates decided to tackle the following challenge:

We were given a dataset of different protocols and the time it took for them to respond to a request. Features we goat included response time, geographical location as well as the exact time. The goal was to predict any latency was to occur. One of the main strategies we wanted to do is find where in the world the request was coming from and then use that to predict the latency based on the weather. If in at a certain time in a certain place in the past, it happened to have stormed, this would have most likely had a role in the latency of different protocols.. The issue is we could not find an API (in a timely fashion) to convert coordinates to weather events. Instead, I prepped the dataset we had by removing outliers, normalizing the data and then splitting it into training and testing sets. We then used a Random Forest model to predict the latency based on the features we had.

so from this :

to this :

We ended trying different models, my friend tried to tackle the problem with an LSTM model using TensorFlow, while I tried to make a TCN model using PyTorch. Unfortunately, we did not have enough time to get a good accuracy on the models we were building. Our R-Squared scores were all over the place on either models so we didn't win any prizes. Still, we got some swag and learned alot which is exactly what we wanted to do.

My repo for my model is found here :

These are some photos I took during the event :)