I am a Data Science and Artificial Intelligence student looking long term to work on machine learning image geolocation to identify images for OSINT teams. I also like to write blog posts and joke articles, here are some of my latest:
You want to speed up your model's training and take advantage of the GPU you paid for, but every guide you find involves a ton of steps, running commands you don't completely trust, or is just so out of date that everything is deprecated. After helping several people navigate this process a friend suggested to me that I should write it up. So here is my no-nonsense guide with all the links to official documentation you'll need. Or skip straight to the GitHub repo here for a template.
Abstract: Identifying where a photo was taken can be achieved by matching the query ground view image to a satellite image of known location. This is done by transforming the images into feature representations and comparing the distance between vectors within the space to find a close match. Historically the number of correct recalls within top 1% of matches was used as a metric. This paper proposes a novel metric, Proportional Search Space Reduction (PSSR), which measures the reduction in the search space of images without the loss of the correct image. Three models were trained and evaluated to show that models with high Recall at 1% do not perform as well in real world applications as the metric may suggest, and proposes the use of PSSR for future research into the problem.
A programming language designed to help you work like a proper Data Science student.
After three years of soaking up all the gems handed down from my professors, I present a guide on how to write exam questions for your students, with examples.
Falling asleep in lectures? Play bingo and be super focused on everything that is happening in the room!
A sudoku solver that uses multiple algorithms to solve sudoku including a heuristic based depth first search and Convolutional Neural Network. The game is available at here as well as a free API with documentation https://sudoku.leondebnath.com/docs and code public on GitHub
An implementation of the game Bamboo (originally designed by Mark Steere) with AI agents to play against. AI algorithms used include: MiniMax, Monte Carlo Tree Search, and a Neural Network. Full paper available here and code public on GitHub
A lightweight Java logging library implemented in a single file to enable students new to java to log with minimal setup.