I attended the workshop Parcimonie, Optimisation et Problèmes Inverses Lyon Saint-Etienne Savoir 2026 in Saint-Etienne (POPILSS 2026). The talks were nice, but one definitely stood out ! The designs from Ammar Mian were awesome. I just checked his personal website and he has some sort of crazy interactive tower now. He also seems to have many other resources that I believe are worth checking out: https://github.com/AmmarMian?tab=repositories
The talk was about semi-positive matrices and how to learn neural networks for these specific structures. I knew a bit of Riemannian geometry and optimization so I didn’t learn much, but the visualizations were very, very, very nice. It reinforced my opinion on the fact that having good supports really improves a talk. Joseph disagreed and thought that some animations were distracting. I could personally have fully visual explanations without formulas and that would be more than enough, walls of equations are not really my thing.