Projects

This project focused on verifying the findings and expanding upon the evaluation and training methods from the paper LICO: Explainable Models with Language-Image Consistency. The main claims are that LICO: enhances interpretability by producing more explainable saliency maps in conjunction with a post-hoc explainability method and improves image classification performance without computational overhead during inference. We reproduced the key experiments conducted by Lei et al. however, the obtained results do not support the original claims....

Using a formal framework of lying and deception, this project done with the AI Safety Hub aimed to investigate the effects of training language models using Reinforcement Learning from Human Feedback (RLHF) in the case where the human is consistently incorrect about a subset of the data. Instead of using a human, we train a language model “judge” to be consistently incorrect about a specific kind of question in a dataset, and the objective is then to see whether the model learns to lie about this kind of question, and whether this lying generalizes to other areas of the dataset....

This was a project in constraint programming that aimed to solve the “Rotating Workforce Scheduling Problem”, as defined in Exact Methods for Extended Rotating Workforce Scheduling Problems. The overall modelling strategy for the model consisted of a viewpoint of assigning an integer shift type to an employee at some day. This can be represented as a 2D array with different employees being indexed through the first dimension and the weekdays being indexed in the second dimension....

In this project I implemented a JSON processor with many (but not all) of the features of jq, an open source command line json processor. Jq itself is written in C++, while this remake uses Haskell. The parsing is based on this surprisingly short functional parsing library. Turns out monads are useful for something. Usage example It also turns out that NASA provides a free API as part of its Sentry system, which tracks objects that are currently projected to impact Earth with a certain probability some time in the future....

For this project I helped in developing a testing procedure and framework for an airport terminal operations simulator. This involved a lot of: refactoring for readability and code quality documenting existing code with JavaDoc documenting the architecturing using UML performing static analysis using checkstyle and PMD creating a testing procedure consisting of micro, meso, and macro level tests implementing unit, integration, smoke, stress, mutation and distribution tests performing test coverage analysis using JaCoCo Top-down view screenshot of the AATOM simulator showing an airport security check

These days SOTA computer vision systems are all based on deep learning and require a lot of data to achieve great results, but what If we don’t have a lot of data, or just want to experience the pain that people went through before they could just slap a neural network on it? Well, in this project we created a car license plate reader based on “traditional” image processing techniques - just with Python and NumPy....

In this project we programmed a simple raytracer that implements hard and soft shadows, reflection, refraction, phong shading, light bloom, motion blur, and anti-aliasing. Rendering is accelerated using bounding volume hierarchies. Some standard meshes rendered using the raytracer are shown below. Rendering the Cornell box using soft shadows and reflections Phong specular highlights Refraction through a translucent sphere Bloom filter for high light intensity on the Stanford dragon Motion blur along a vector towards the dragon...

Yeah, the name is a bit dramatic… NDW is a fancy-looking game I made with a teammate that uses websockets to connect you to another player and lets you play connect 4 in real time. It was one of my first projects, so don’t expect amazing software quality :) Splashscreen of NDW Perspective of one player Perspective of the opposing player Win screen with winning row of chips highlighted