Discovery of symbols in design

In this work, we propose a mechanism for learning grounded symbols based on design experience as acquired by a CAD system. We discover the functionally feasible regions FFRs lying along lower dimensional surfaces or manifolds, reflecting latent interdependencies between design variables sometimes called chunks constitute the intermediate step between raw experience and the eventual symbol that arises after these patterns become stabilised through communication. We search this lower-dimensional manifolds using either linear or nonlinear techniques, and demonstrate low-dimensionality at three stages of an 'infant designer': a learning image schemas for fits and clearances using linear algorithms; b universal motor, 2-parameter - learning a non-linear manifold; c universal motor, 8-parameter, reducing to two parameters or chunks using both linear and nonlinear mappings.