Louis Bethune
Louis Bethune
As noticed in [this discussion](https://github.com/google/jaxopt/discussions/252#discussioncomment-3034571) the default hyper-parameters for EqualityConstrainedQP might cause divergence in some circumstances for [projection_affine_set](https://github.com/google/jaxopt/blob/fdaf31d7e369c0bcdf46705c86d068d56f0d3d90/jaxopt/_src/projection.py#L278) It would be nice to let the user access those hyper-parameters. Similarly,...
## What's new ? Two new classes: `NMF` for **Non Negative Matrix Factorization**. This non-convex problem is NP-hard [3], but it is bi-convex: Solves :: min_{H1, H2} 0.5 * ||Y...
It will be easier to discuss this on github rather than internal Google doc. ## The issue Current state of API: | Function | Without ridge | With ridge r...
Add a new function that returns a point lying in the intersection of convex sets. This is not a true projection, but not too far from being one, so it...
## The problem (issue #342) Following the discussion of #342 it has been noticed that the default zero initialization for dual variables was harmful for convergence, specially when primal variables...
Dear authors, Thanks for this work According to the paper, Appendix F.1 in page 25: "To enforce independence of the step-size and stochastic gradient, we perform a backtracking line-search at...
Currently the `requirements` file force the installation of the two frameworks. Maybe two files would be better since a typical user is only interested in one of the two implementations.