The University of British Columbia

Purpose

Many design problems involve geometric relationships or shapes. In these problems, the central issue can often be formulated to computing optimal shapes or configurations subject to a set of constraints. Often numerical simulations can be performed based on physical laws to virtual test the effectiveness of a design. Recent advances in machine learning allow for a data-driven solution: a neural network can be trained using pre-computed simulation data and then use the neural network as an online simulator. This project investigates geometric techniques for representing and processing shape data for machine learning and computing optimal shapes for design.