Performance estimation: SDPs were auto-proving algorithm convergence long before LLMs
burny_tech · x · 2026-10-08
Responding to buzz about a theoretical breakthrough on gradient descent and matrix multiplication, this thread notes the idea isn't new: convex optimization has long used semidefinite programs to automatically generate proofs that an algorithm converges — so-called performance estimation problems, cf. Taylor's work.
More from Research
- Medical AI benchmark builder: his first ECG dataset had answers printed on images — MaziyarPanahi · 2026-10-08
- Learned latent dynamics roll forward from few observations to recover hopper state with uncertainty — hisspikeness · 2026-10-08
- Adding a pixel-reconstruction loss makes latents unable to decode hopper state — hisspikeness · 2026-10-08
- New theory shows SSL separates stochastic signals from nuisance via MI maximization plus distribution matching — hisspikeness · 2026-10-08
- The conundrum of latent prediction: signals themselves are stochastic, so what to keep? — hisspikeness · 2026-10-08
- Hippocampal cognitive maps support language learning, not just spatial navigation — abenitezburraco · 2026-10-08