Active learning splits composition from processing in optical materials optimization
bravo_abad · x · 2026-09-23
A paper on circularly polarized white-light materials shows why one optimizer isn't always enough. Chemical composition mainly controls emitted color, while film thickness, stretching and twist control circular polarization strength — different physics, so the authors run two separate optimization loops.
- Composition loop: a Wasserstein autoencoder builds a smoother latent space from the small experimental dataset, with TabPFN ranking candidate compositions
- Processing loop: Gaussian-process Bayesian optimization tunes fabrication parameters for the best compositions
The takeaway: when different variables control different properties, decoupling them into specialized loops beats forcing a single model to optimize everything at once.
More from Research
- Block-triangular joint drifting enables one-step generative surrogate models for stochastic trajectories — chaumian · 2026-09-23
- Raw LLM probabilities aren't enough for decisions — calibration matters, researchers argue — PMinervini · 2026-09-23
- Parallel search blunts Grover's speedup, making AES-256 harder to break than thought — Jsevillamol · 2026-09-23
- Hodge and Yang-Mills still lack full Lean formalizations, making near-term solutions less likely — Jsevillamol · 2026-09-23
- Steve Hsu: AI will push math frontier far beyond human minds, compression defines 'human math' — burny_tech · 2026-09-23
- AI boosts science productivity — but mostly spawns spam papers, says Ehud Reiter — EhudReiter · 2026-09-23