Freelance data scientist · All case studies

A checkerboard, three generative stories.

GANs, denoising diffusion (DDPM), and implicit diffusion (DDIM) trained on the same 2D synthetic target. Energy Distance and 2-Wasserstein as the scoreboard — not a pretty GIF.

Client: Generative benchmark. Built by Dilshad Raza.

2D synthetic densities are where generative methods should be comparable. A GAN that mode-collapses and a diffusion that oversmooths can both make a nice screenshot. The work needed the same target, the same metrics, and enough runs to talk about error, not vibes.

I trained and evaluated GANs, DDPM, and DDIM on the checkerboard distribution and reported Energy Distance and 2-Wasserstein so a collapse or a blur showed up as a number, not a thread of cherry-picked frames.

Hire Dilshad Raza for similar freelance data science, machine learning, and AI automation in the UK, United States, and Australia.