The world is being quietly rearranged by people who write very long documents.


The title they went with NeST-BO: Fast Local Bayesian Optimization via Newton-Step Targeting of Gradient and Hessian Information Noisy translates that to

Faster optimization for complex problems using curvature learning


Researchers developed a method that learns the shape and slope of expensive computational problems more efficiently, allowing algorithms to find good solutions in fewer tries. This matters because many real-world problems—drug discovery, materials design, engineering tuning—are too costly to try many times, so smarter guessing saves money and time.
The method reduces what was a quadratic scaling problem (cost grows with dimension squared) to a linear one by working in lower-dimensional projections, which means optimization of high-dimensional problems now becomes practical rather than computationally prohibitive.

If you insist
Read the original →