Hadi Dolatabadi — opinions, explainers, and the occasional walk-through.

Decentralised training is how open-source models become financially meaningful. I work on the parts that break: pretraining, security, privacy.

Research scientist at Pluralis Research in Melbourne. I'm statistical-minded and opinionated, usually in defence of whatever the numbers actually support. Most of my time goes to optimisers and architectures friendly to decentralised training, with a pinch of adversarial ML to keep things safe.

I did my PhD at the University of Melbourne on adversarial machine learning under Sarah Erfani and Christopher Leckie, with an Amazon internship along the way. After a postdoc with the ARC Centre of Excellence for Automated Decision-Making & Society and a year and a half at Oracle building LLM-based applications for healthcare, I joined Pluralis.

This site is where I write in public. Expect opinionated takes on what works, reading notes when a paper actually teaches me something, and the occasional interactive essay when prose runs out of room.