Using LLMs to summarize GitHub issues
A learning exercise on using large language models (LLMs) for summarization. It uses GitHub issues as a practical use case that we can relate to.
About
I’m a software engineer at MathWorks. More about my professional life on LinkedIn.
In December of 2020 I completed a Computer Science master’s degree at Florida Atlantic University, focusing on machine learning. The thesis is available here. I’m now a Ph.D. candidate, focusing on machine learning, continuing the work I did for the master’s degree. My master’s and Ph.D. projects are on GitHub.
Publications
- Master’s thesis.
- Code for assignments, research projects, and blog posts are on GitHub.
Papers and articles where I’m the main author:
- Machine learning for predicting opioid use disorder from healthcare data, a systematic review. The accepted manuscript is available here (free to access).
- Assessing Methods and Tools to Improve Reporting, Increase Transparency, and Reduce Failures in Machine Learning Applications in Health Care, on the RSNA Radiology:Artificial Intelligence journal (free version available on the NIH Library of Medicine site). This article is based on the master’s thesis.
- Dataset datasheet for CheXpert, in collaboration with the Stanford CheXpert team.
- Dropout vs. batch normalization. The free version of the paper is available here and the code is on GitHub.
Papers and articles where I’m a collaborator:
- Automated patient localization in challenging hospital environments.
- Deep learning for edge devices.
- On the use of colormaps in the context of post-hoc explainable AI methods, an abstract session in SIIM 2022.
- IEEE ICMLA data science tutorial notebooks, the practical part of Dr. Marques’ The Data Science landscape: foundations, tools, and practical applications session.
More publications and patents on Google Scholar.
Older things
I used to be more active on Stack Overflow. Nowadays I help a bit with questions and answers moderation.
Earlier I wrote a few pieces on programming topics on Blogger. I switched to GitHub Pages (this blog) to simplify my workflows because most of my work is already on GitHub.
Recent posts
Writing good Jupyter notebooks
How to write well-structured, understandable, flexible, and resilient Jupyter notebooks.
Vision transformer properties
Vision transformers are not just a replacement for CNNs and RNNs. They have some interesting properties.
Understanding transformers in one morning
Transformers: from zero to hero in one morning (or at least know enough to discuss transformers intelligently and apply them to your projects).
Applications of transformers in computer vision
The evolution of transformers, their application in natural language processing (NLP), their surprising effectiveness in computer vision, ending with applica...
Machine learning interpretability with feature attribution
A review of feature attribution, a technique to interpret model predictions. First, it reviews commonly-used feature attribution methods, then demonstrates f...
An overview of deep learning for image processing
Deep learning (large, multi-layered neural networks) have been successfully applied to computer vision tasks. This article reviews its origins, the evolution...
Exploring SHAP explanations for image classification
How to interpret predictions of an image classification neural network using SHAP.
Machine learning, but not understanding
In the expression ‘machine learning’, are the machines actually learning anything? Let’s explore what ‘learning’ means for machine learning, guided by Melani...
What is “model accuracy”, really?
A model’s accuracy is an incomplete view of of the model’s performance. This article shows how it can be misleading.
Fairness in machine learning: a reading list
A list of resources to understand concepts and applications of fairness in machine learning (ML).
Explainability: end-users considerations
If we assume that explaining to the end-users how a machine learning (ML) model makes its predictions increases their trust on that model, the question is th...
Would you trust AI to do [X]?
Exploring ‘robustness’ as a factor to trust AI products, with examples of how difficult it is to create robust AI products.
Bias in data science and machine learning
Of all the problems that may crop up in the machine learning lifecycle (acquire data, train a model, test the model, deploy, and monitor), biased data is the...
Artificial intelligence and jobs
A review of articles that explore the effect of AI on jobs, based on an article from The Economist.