Sean Hanna

Automated Representation of Style by Feature Space Archetypes: Distinguishing Spatial Styles from Generative Rules

Style is a broad term that could potentially refer to any features of a work, as well as a fluid concept that is subject to change and disagreement, yet approaches to representing it too often seek either a pre-defined set of generative rules or list of measurable features. Instead, a general and flexible method of retrospectively and automatically representing style is proposed based on the idea of an archetype, to which real designs can be compared, and tested with examples of architectural plans. Unlike a fixed, symbolic representation, both the measurements of features that define a style and the selection of those features themselves can be performed by the machine, making it able to generalise a definition automatically from a set of examples. This process is implemented in analysis, and coupled with a generative algorithm to produce plans in a learned style.

Inductive machine learning of optimal modular structures: Estimating solutions using support vector machines

Abstract Structural optimization is usually handled by iterative methods requiring repeated samples of a physics-based model, but this process can be computationally demanding. Given a set of previously optimized structures of the same topology, this paper uses inductive learning to replace this optimization process entirely by deriving a function that directly maps any given load to an optimal geometry. A support vector machine is trained to determine the optimal geometry of individual modules of a space frame structure given a specified load condition. Structures produced by learning are compared against those found by a standard gradient descent optimization, both as individual modules and then as a composite structure. The primary motivation for this is speed, and results show the process is highly efficient for cases in which similar optimizations must be performed repeatedly. The function learned by the algorithm can approximate the result of optimization very closely after sufficient training, and has also been found effective at generalizing the underlying optima to produce structures that perform better than those found by standard iterative methods.

Beyond simulation: designing for uncertainty and robust solutions

Simulation is an increasingly essential tool in the design of our environment, but any model is only as good as the initial assumptions on which it is built. This paper aims to outline some of the limits and potential dangers of reliance on simulation, and suggests how to make our models, and our buildings, more robust with respect to the uncertainty we face in design. It argues that the single analyses provided by most simulations display too precise and too narrow a result to be maximally useful in design, and instead a broader description is required, as might be provided by many differing simulations. Increased computing power now allows this in many areas. Suggestions are made for the further development of simulation tools for design, in that these increased resources should be dedicated not simply to the accuracy of single solutions, but to a bigger picture that takes account of a designs robustness to change, multiple phenomena that cannot be predicted, and the wider range of possible solutions. Methods for doing so, including statistical methods, adaptive modelling, machine learning and pattern recognition algorithms for identifying persistent structures in models, will be identified. We propose a number of avenues for future research and how these fit into design process, particularly in the case of the design of very large buildings.

Defining implicit objective functions for design problems

In many design tasks it is difficult to explicitly define an objective function. This paper uses machine learning to derive an objective in a feature space based on selected examples of previous designs, thus implicitly capturing the features that distinguish that set from others without requiring a predetermined measure of fitness. A genetic algorithm is used to generate new designs, and these are shown to recognisably display the appropriate features. It is demonstrated that the range of relevant features and optimal solutions is easily varied in proportion to the examples selected to define the objective. Methods for improving the function for GA search are discussed.

A Redefinition of the Paradox of Choice

Barry Schwartz defined the paradox of choice as the fact that in western developed societies a large amount of choice is commonly associated with welfare and freedom but too much choice causes the feeling of less happiness, less satisfaction and can even lead to paralysis. The paradox of choice has been recognized as one of the major sources of mass confusion in context of the B2C online mass customization. We propose to redefine the paradox of choice with an emphasis on the meaning of choice in conjunction with the amount of available options, rather than just the quantity of choice. We propose that it is the lack of meaningful choice, rather than an overwhelming amount of choice, that can cause customers’ feelings of decreased happiness, decreased satisfaction and paralysis. We further propose that since users themselves are often not able to explicitly define what constitutes a meaningful choice, the task they face belongs to the category of ill-defined problems. The challenge for mass customization practitioners is thus not to limit the scope of choice, as has been suggested in previous literature, but to provide users with choice that is relevant to them. We further discuss two computational approaches to solving problems related to the redefined paradox of choice in the context of the B2C mass customization. The first is based on recommender systems and the second is an implementation of artificial selection in genetic algorithms. We present findings of an empirical comparison of genetic algorithm and parametric product configurators. We find that the genetic algorithm tools, which allow users to move through a solution space by recognition of meaningful options rather than their definition, appear to be more popular among the users when it comes to browsing through solution spaces with larger number of dimensions.

Representing style by feature space archetypes

Style is a broad term that could potentially refer to any features of a work, as well as a fluid concept that is subject to change and disagreement. A similarly flexible method of representing style is proposed based on the idea of an archetype, to which real designs can be compared, and tested with examples of architectural plans. Unlike a fixed, symbolic representation, both the measurements of features that define a style and the selection of those features themselves can be performed by the machine, making it able to generalise a definition automatically from a set of examples.

A Generic Shape Grammar for the Palladian Villa, Malagueira House, and Prairie House

Shape grammars are formulations consisting of transformation rules that describe design. Previous studies have focused on recreating the style of family-related solutions. This study does not aim to recreate a specific architectural style but is part of wider research aimed at inferring shape grammars. It is believed that more than one grammar can be developed for the same style, but no one has ever demonstrated this possibility. In addition, no one has ever developed a grammar that can describe more than one style. The aim of this work is to demonstrate both possibilities. Firstly, it proposes a shape grammar that can produce three different design styles, and, secondly, it uses a process that is distinctively different from other tested examples yet still produces the same corpus of designs. It also enables a new corpus of designs to be produced, which had not been possible using the previous (or original) grammars. A selected case study of three grammars, namely for Palladian, Prairie and Malagueira houses, allowed for comparison and observation of the different processes and shape rules and for a new set of rules to be proposed, combined in a shape grammar. This was followed by the recreation of a new subdivision type of grammar with a top–down approach and a set of generic design rules. The result is a generic shape grammar that enables three different house styles to be designed from the same formulation.

A representational scheme for the extraction of urban genotypes

A representational scheme is described for cities, which uses the spectrum of the graph derived from a network of streets. This is of a sufficiently high dimensionality both to capture information of the city structure and to allow different representations of urban types to be extracted from it. It is proposed that a machine can extract the ‘genotype’ description that classifies a given group of cities. Results demonstrate that these capture morphological relationships between cities, and reveal correlations between these and a city’s geographical location. This has implications for our understanding of design processes and the modelling of creativity, in that the final representation can be made autonomously by the computer, rather than predefined by a priori standards.

Blurring the boundaries between actuator and structure: investigating use of stereolithography to build adaptive robots

The aim of this work is to investigate whether stereolithographic models, combined with shape memory alloys, can be used to create robotic structures that act both as support structure and actuator. In order to test this, a structural topology was evolved and subsequently, through a deterministic process, optimised to withstand predetermined forces similar to those possibly encountered in a robot. The results show that by using this two-stage process, structures can be designed and built that satisfy these requirements.

A new palladian shape grammar: A subdivision grammar as alternative to the palladian grammar

The following paper describes a shape grammar that recreates Palladian villas. A Palladian grammar was originally proposed by Stiny and Mitchell. However, this alternative grammar uses different parametric shape rules and methodology to test the hypothesis that different grammars can generate the same corpus of designs. The formalism is then implemented using a computerised design tool. The grammar includes subdivision rules that allow for a more economical formulation. A new corpus of solutions is explored and the derivation is compared with the original Palladian grammar. The project is part of wider research aimed at formulating a generic housing grammar.