Burke O’Brien


 
I’m a data scientist interested in applying rigorous machine learning methods to tell important and compelling narratives. I’m currently interning with the Analytics Innovation team at Altman Solon, a leading strategy consulting firm, where I support their analytics and machine learning capabilities.

With my background in academia and public policy research, I offer a unique perspective and range of experiences applying statistical and machine learning methods to real-life problems.

Outside of work, I enjoy crosswords, playing games with friends, and seeing the city from my bike. Ask me about Boston sports teams, my love of chess, or why I will never again operate a coffee-shop.

 
Education

2022-2023
Brown University
Master of Science in Data Science 

2015-2019
Brown University
Bachelor of Arts in Economics + International Relations
 
Let’s Connect
Email
LinkedIn
Resume
Let’s play chess!
Lorem Ipsum

Hi!

Email
Instagram
Twitter

Automatic Music Genre Classification


Music genre classification is a challenging task for deep-learning models; there is high variability in music where tempo, rhythm, and melody can differ within genres. There is also the question of how to best represent audio data for this task. Transformers, which have been shown to be highly effective in Natural Language Processing (NLP) tasks, could help learn the long-term dependencies present in audio files that contribute to the human construct of music genre. 

The model we implemented uses a stack of Transformer blocks to learn long-term dependencies in our song representation, and then an MLP stack outputs genre probabilities.

A large component of this project is interpretability- we analyzed how our model makes predictions by mapping the Transformer attention, and by using the “XpLique” package (developed at Brown!). This analysis revealed interesting relationships between music genres, as well as highlighted how this popular dataset may not be well suited for this task.

The goal of this project was to provide greater insight into the challenges and opportunities of music genre classification and shed light on the interpretability of deep-learning audio models.


Read more about this project on this Devpost

Tools:
→ Python
   - Tensorflow
   - Keras
   - Librosa
   - Xplique
→ Deep Learning
   - ResNets
   - Transformers/Attention
   - Image preprocessing
   - Classification pipelie