New roles for machine learning in design

Computer based pedestrian landscape design using decision tree templates

“…A study on the role of machine learning for well defined design problems and their limitations when applied to real world tasks has been discussed in Ref. [1].…”

“…From the above example, we can identify the following six steps for applying machine learning (ML) techniques in design [1]: (i) formulation of the learning problem in a particular design context, (ii) preparing the inputs, (iii) developing a representation for the input information, (iv) selecting a learning algorithm, (v) selecting operational parameters, and, (vi) analyzing results. Fig.…”

“…In the context of design, machine learning has been primarily used to extract symbolic knowledge from previous design data using inductive learning approaches [1], recall and reuse of specific previous design experience through casebased reasoning [2], and addressing combinatorial optimization problems in design with genetic algorithms [3]. A range of design problems from simple discrimination of feasible designs from infeasible ones [4] to complex tasks like optimizing architectural floor plans [3], completing partial room designs [5], and generating bridge designs [6,7] have already been investigated using these approaches.…”

Facilitating design learning through faceted classification of in-service information

“…Methods generally classified under text mining can help towards automating the identification of concepts from the descriptions and the generation of rules/constraints for those concepts [37,38]. Text mining is useful as the classification is constructed using literary warrant, i.e.…”

“…This construction does not have to be conducted for all documents, a subset of documents may be interrogated for this purpose (called the training set). Other established techniques exist, for example it is possible to automatically parse the documents to extract all significant terms (for example, by comparing the prevalence of certain terms within a document as compared to the prevalence within the overall corpus, referred to as "term frequency -inverse document frequency" [37,38]) and to organise those key terms to form concepts. Methods for assisting concept identification will be discussed for the case study in later sections although a controlled vocabulary can greatly assist in this task.…”