A review is given of approaches to growing neural networks and electronic circuits. A new method for growing graphs and circuits using a developmental process is discussed. The method is inspired by the view that the cell is the basic unit of biology. Programs that construct circuits are evolved to build a sequence of digital circuits at user specified iterations. The programs can be run for an arbitrary number of iterations so circuits of huge size could be created that could not be evolved. It is shown that the circuit building programs are capable of correctly predicting the next circuit in a sequence of larger even parity functions. The new method however finds building specific circuits more difficult than a non-developmental method.

Traditional evolutionary circuit design does not scale well to large, complex problems. Nature solves the scalability problem by using a complex mapping implicit in the process of biological development. By modelling this process we aim to improve scalability in evolutionary circuit design. Here we extend our earlier work [1] by demonstrating that evolution can learn and encode useful circuit design abstractions in a developmental process. We go on to present enhanced models of development with improved intercellular communication and show how this improves their ability to generate circuits.

The scalability problem is a major impediment to the use of hardware evolution for real-world circuit design problems. A potential solution is to model the map between genotype and phenotype on biological development. Although development has been shown to improve scalability for a few toy problems, it has not been demonstrated for any circuit design problems. This paper presents such a demonstration for two problems, the n-bit adder with carry and even n-bit parity problems, and shows that development imposes, and benefits from, fewer constraints on evolutionary innovation than other approaches to scalability. 1.

The effect of phenotypic complexity on distance correlation plots is investigated for two developmental mappings, a mapping based on L-systems, and a 2D cellular automata mapping. Our treatment of complexity is based on the theory of Kolmogorov complexity. A new genotype sampling algorithm called Crossection Walk is introduced.

Abstract The fractal protein is a new concept for improving evolvability, scalability, exploitability and providing a rich medium for evolution. Here the idea of fractal proteins is introduced, and a series of experiments showing how evolution can design and exploit them within gene regulatory networks is described. 1

In this articl e, we introduce the approach to the real ization of ontogenetic devel2 ment and fau l tol02( ce that wil be impl emented in the POEtic tissue, a novel reconfigurabl e digital circuit dedicated to the real2H) ion of bio-inspired systems. The model ization in el ectronic hardware of the devel opmental process of mu l i-cel-- ll biolWH(:D organisms is an approach that cou l become extremel useful in the impl ementation of high l complm systems, where concepts such assel2 organization and faul t tol))-- ce are key issues. The concepts presented in this articl e represent an attempt at finding a useful set of mechanisms to al l ow the impl ementation in digital hardware of a bio-inspired devel2 mental process with a reasonab l overhead. 1

The introduction of a genotype-phenotype map modelled on biological development can potentially improve the scalability of evolutionary algorithms. Previous work by Gordon and Bentley demonstrated that such a model can be used to evolve patterns that map to useful but small phenotypes. This paper uses the same model to generate much larger patterns covering arrays of up to 64x64 cells. The results show that the model’s performance is generally comparable to similar development-based systems [12, 14], and with some measures outperforms them. Additionally the inherent biases of the model are explored, such as the need to use symmetry-breaking initial conditions which some other models do not require. This exploration yields a set of guidelines that suggest what kinds of problem the model is suited to exploring.

Artificial Development is a promising approach to evolutionary design optimization inspired by biological development. However, there is still no consensus as to which problem classes this approach has a clear advantage over classical direct encodings. We attack this problem by introducing the concept of fitness function modifiers based on complexity. Our results indicate that using these modifiers, we are able to discriminate a developmental mapping from a direct encoding with respect to their e#ciency at solving classes of problems defined by the fitness modifiers.

This paper deals with the evolutionary design of programs (constructors) that are able to create (n+2)-input circuits from n-input circuits. The growing circuits are composed of polymorphic gates considered as building blocks. Therefore, the growing circuit can specialize its functionality according to environment which is sensed through polymorphic gates. The work was performed using a simple circuit simulator. We evolved constructors that are able to create arbitrarily large polymorphic even/odd parity circuits and polymorphic sorting networks.

In this article, we try to analyze the requirements of developmental processes from the perspective of their implementation in digital hardware. After recalling the motivations for such an implementation, we concentrate separately on the two mechanisms (cellular division and cellular differentiation) that are exploited by biological systems to realize development. We then describe some of the current and projected solutions to implement such mechanisms in hardware, and conclude by analyzing the most interesting features of developmental approaches. 1