Simon Vary retweeted
📄 Masked Language Flow Models (MLFMs)
We introduce a method to scale flow-based language models to reasoning and instruction-following tasks.
Joint work with @kiaashour @leoeleoleo1 @simonvary @prebeschini
🔗 kiaashour.github.io/blog/mas…
1/6
Simon Vary retweeted
[Paper alert] "Optimization without Retraction on the Random Generalized Stiefel Manifold"
Introducing a cheap method to optimize functions on a *stochastic* manifold, defined by an expectation constraint.
arxiv.org/abs/2405.01702
See you in ICML :)
Simon Vary retweeted
The Learning Theory & Statistics Optimization Reading Group meeting info:
📕Optimization with retraction on the random generalized Stiefel manifold.
🎙️Speaker: Šimon Váry
⌚Time: 4.15pm – 5.15pm
📍Location: Open Research Zone, Dept. of Stats
Open to #oxforduni UG, PGT, + PGR – at Oxford, England
ALT A dark image with grey, light blue and white text on it providing information about the Reading Group Session. The image reads: “Learning Theory and Statistics Optimization Reading Group. Optimization with retraction on the random generalized Stiefel manifold. Speaker: Šimon Váry. Time: 4.15pm – 5.15pm. Date: Thursday 7th March 2024. Location: Open Research Zone, Department of Statistics (in person only).”
Simon Vary retweeted
Preprint alert!! arxiv.org/abs/2303.16510
"Infeasible Deterministic, Stochastic, and Variance-Reduction Algorithms for Optimization under Orthogonality Constraints"
w. P.-A. Absil, S.Vary, and B.Gao.
1/2
