Jean Lucien Randrianantenaina

HASTE: Hardware-Aware Dynamic Sparse Training for Large Output Spaces

Nasib Ullah, Jinbin Zhang, Jean Lucien Randrianantenaina, Erik Schultheis, and Rohit Babbar

Extreme multi-label classification requires learning over output spaces with millions of labels, making the output layer a memory and compute bottleneck. We introduce group-shared fixed fan-in sparsity with hardware-aware CUDA kernels, achieving up to 4.4x and 25x speedups in the forward and backward passes while matching dense-model accuracy.

There Is No Carmichael Number of the Form $2^np^2+1$

Florian Luca and Jean Lucien Randrianantenaina

In this paper, we prove that there is no Carmichael number of the form $2^np^2+1$ with some integer…

Fermat’s Last Theorem, with $n=4$

A presentation of the $n=4$ case of Fermat’s Last Theorem, centered on the classical infinite descent argument and the way it rules out non-trivial integer solutions.