Q

Q

We are a research lab focused on understanding and solving generalization. Our goal is to build learning algorithms beyond gradient descent where generalization scales optimally with compute, even with minimal data. Read more about our research direction.

We're hiring researchers–reach out at research@qlabs.sh

Progress

Decoupling Search and Learning

Our first paper on search. We decompose training into search in representation space and learning in parameter space. It shows tractable evolutionary search can guide gradient-based learning—approaching SGD while following different optimization paths.

Generalization From First Principles

Our core framework for solving generalization. We describe how to combine search and generalization theory to create algorithms that approach optimal learning—where generalization scales with compute even on minimal data.