Researcher shows EM M-step derivations for factor analysis can all be made convex
caglar_ee · x · 2026-11-09
Researcher caglaree shares a technical observation: M-step derivations for factor analysis, linear dynamical systems, and mixtures of factor analyzers are usually framed as nonconvex optimization problems in textbooks and papers.
Once he reformulated one of them as a convex optimization problem, the rest followed easily — and it turns out all of them are convex. The reframing offers a useful lens for deriving and understanding EM-type models.
Related event: Researcher Shows Classic EM M-step Derivations Rest on Non-convex Problems(2 posts)→
More from Research
- PCBRouteAgent: multimodal LLM agent routes 84/86 boards vs baseline 10/86 — Yue Zhang · 2026-11-10
- NVIDIA's CPU-ARC: reasoning caches 16–291x cheaper with frontier-level performance — nvidia · 2026-11-10
- Not even wrong: why AI won't conquer theoretical physics — calabi_and_yau · 2026-11-10
- First blinded RCT of AI-generated full echo reports presented at AHA — anshulkundaje · 2026-11-10
- Dev claims Gaussian Splats are a dead end for interactive 3D experiences — iandanforth · 2026-11-10
- MuJoCo 4.0 and Playground 1.0 ship: dented clay, creased paper, 6-9x faster grasping — DynamicWebPaige · 2026-11-10