Zak Buzzard
Hi! I'm a University of Cambridge computer science Master's graduate doing ML research, mainly computer vision related.
Also applying AI to videogames at ZogSoft!
Hi! I'm a University of Cambridge computer science Master's graduate doing ML research, mainly computer vision related.
Also applying AI to videogames at ZogSoft!
Efficient whole slide image analysis using a hierarchical transformer and iterative patch zooming.
A novel joint value function decomposition for multi-agent reinforcement learning. Improved expressivity, but requires maximisation through dynamic programming.
An implementation of a token-level diffusion language model, with several novel techniques for controlling the quality-diversity trade-off.
Custom multi-agent reinforcement learning environment able to scale to many agents. Designed for use with PettingZoo, and features in their third party environment list.
A Tkinter app which uses repeated outpainting to give an infinite seamless scrolling view. Can also be used to just create really long images.
Generate random animations where each frame is an image variation of the previous frame. Uses CLIP guidance and a model finetuned to condition on image embeddings.
Improving Gaussian splatting reconstructions with synthetic data from a fine-tuned diffusion model.
Zero-shot panorama generation from pre-trained GANs by projecting features between cameras for multi-view consistency. Can also be used as a seamless sphere texture generator.
A Gradio app for creating weird GAN-based animations, with support for BigGAN and StyleGAN.
A tiny diffusion model implemented + trained from scratch on 32x32 pixel art characters, with native transparency support.
A tool for explaining strange Python objects given to you in interactive sessions.