PhD student, University of Wisconsin - Madison
1 paper at NeurIPS 2025
We show that linear layers, the primary player in large language models, can be composed from fundamental algebraic primitives with exponentially less parameters.
