benchmark_bilevel
benchmark_bilevel copied to clipboard
Benchmark for bi-level optimization solvers
This PR aims at simplifying the benchmark and making it more user-friendly. This is mostly done by removing the coexistence of the three implementations for each solver (pure `Python`, `Numba`and...
This PR add the 20news dataset for multilogreg task. We work with BCOO matrix for the moment because these are the only class of sparse matrix implemented in JAX that...
This adds a distillation experiment, where the inner problem is training a nn on some samples; those samples are the outer variable, and the outer problem is validation on mnist....
We should add a description of the quadratic task in the README Fixed in #40
We have added three new algorithms: SPABA (Chu et al., ICML 2024), SFFBA, and MSEBA. For further details, refer to the $\texttt{spaba.py}$, $\texttt{sffba.py}$, and $\texttt{mseba.py}$ files located in the $\texttt{solver}$...