I'm Kegan.

Welcome to my website. Here you can see some of my projects, learn about more about me and get excited about the cool things I can build for you.

This a demo I've been working on to improve GPS tracking. The dots above move about the area randomly while a neural network learns how to predict their next positions from their speed and direction. The squares represent where the network believes the dots to be based on the last location, direction and speed. To represent more realistic data there's also 20% random noise in both direction and speed.

This network graph shows the activations and weights of the network learning to follow dots in the previous section. The edges represent the weights and the node color represent the layer activation. Low values appear red and high values appear purple.

I made a GAN (Generative Adverserial Network) to generate human faces. Using the popular celebrity face dataset CelebA I trained a discriminator to detect generate images and a generator to create images. The two fought against each other for thousands of epochs (about 2 days) and here are some of the final samples:

Not all of the generated images were winners. Here's some of my favorite eldritch-horrors from training:

I'm constantly experimenting and researching new Machine Learning methods. You can checkout some of my work HERE. I've played around with a bunch of different modules and frameworks, but nothing beats building your own model and setting out on a problem.

I love building websites. I built this website myself from scratch. Yeah, there are way easier ways to make websites these days, but I enjoy building them. You can checkout the code here: https://github.com/Aabglov/Aabglov

I'm not all business though. My friends and I LOVE Nicolas Cage. We're watching his movies. All of them. In chronological order of release. Check out what we're watching: this week