About
Independent RL Lab.
I use specified environments to study how learned policies change with dynamics, rewards, exploration, and information.
THE NAME
IRL stands for independent reinforcement learning. The lab is separate. The hyphen leaves the other reading in place.
I returned to reinforcement learning after time away and wanted better machinery than a folder of isolated notebooks. The result is both a software project and a sequence of public experiments: environments with known mechanics, exact solvers where possible, matched comparisons, versioned artifacts, and notes that distinguish what the model predicts from what the data supports.
The current experiments focus on finite environments where the transition model can often be solved numerically. They concern specific learning mechanisms and do not make claims about intelligence in general.
I am Patrick Starvaggi. This is an independent project, not an institution, company, or academic lab.