Darby Losey, PhD


Machine learning researcher and engineer

Darby’s Paper Lab Interactive essays on the papers behind modern sequence modeling Hands-on demos and deep-dive tutorials, starting with the memory theory behind S4 and Mamba. Draw a signal and see how a handful of numbers can hold an entire history. Explore Darby’s Paper Lab →
Darby Losey

Welcome. I am currently a senior software algorithms engineer at TDK USA Corporation, where I work on multimodal language modeling.

My own research interests center on subquadratic neural network architectures, particularly state space models. I am drawn to the question of how sequence models can capture long-range structure without paying the quadratic cost of attention, and to what the inductive biases of these architectures reveal about the sequences they are asked to model.

I hold a joint PhD between the Machine Learning Department and the Program in Neural Computation at Carnegie Mellon University, where my research focused on neural manifolds and the mechanisms underlying multi-task learning and catastrophic forgetting. I also hold a master's degree in machine learning from Carnegie Mellon.

Prior to this, I worked at a startup called Neubay, building deep learning models. I did my undergraduate at the University of Washington in computer science and neurobiology. During this time, I conducted research developing new methods for brain-computer interfacing.