Dimos Tsouros

We propose a highly efficient loss function that exploits the geometry of linear optimization to enable fast solver-free training without compromising decision quality.