Efficient probabilistic grammar induction for design

By Mark E. Whiting, Jonathan Cagan, Philip R. Leduc. Artificial intelligence for engineering design, analysis and manufacturing, 2018

296 citations recorded by the source corpus.

Abstract The use of grammars in design and analysis has been set back by the lack of automated ways to induce them from arbitrarily structured datasets. Machine translation methods provide a construct for inducing grammars from coded data which have been extended to be used for design through pre-coded design data. This work introduces a four-step process for inducing grammars from un-coded structured datasets which can constitute a wide variety of data types, including many used in the design. The method includes: (1) extracting objects from the data, (2) forming structures from objects, (3) expanding structures into rules based on frequency, and (4) finding rule similarities that lead to consolidation or abstraction. To evaluate this method, grammars are induced from generated data, architectural layouts and three-dimensional design models to demonstrate that this method offers usable grammars automatically which are functionally similar to grammars produced by hand.

Subfields: Architecture, Building and Construction, Mechanical Engineering

View Paper: https://www.semanticscholar.org/paper/39593dc89d2c3d8c3e61b4a042f8e753b37e7106