Figure 13 - uploaded by Julian Francis Miller
Content may be subject to copyright.
Source publication
Intrinsic evolution is often limited to using standard electronic components as the media for problem solving. It has been argued that because such components are human designed and intentionally have predictable responses, they may not be the optimal medium to use when trying to get a naturally inspired search technique to solve a problem. Evoluti...
Context in source publication
Context 1
... response. A voltage was applied to the EM and ramped from 0V to 3V over a period of approximately 2ms, figure 6. The response from the EM was sampled over this period using the analogue inputs of the PC. The fitness function then parsed the sample and averaged over a number of contiguous samples to produce an expected value for the result in the middle of a small window (of length w, where w=3), figure 7. If the LC response was linear the expected voltage value would be the same as the measured output value. If the response is non-linear (e.g. a step change) then the expected value will be different from the recorded value. The total fitness of the response was calculated as the sum of the differences of the expected value for a sample and its actual value. As a side effect, by averaging several values to find the expected response the sample was ”smoothed” and noise reduced. Let S be a set of L samples, the input and output samples are represented as S in and S out . The j th element of the set is S [ j ] . Several different but common types of response were observed. Sometimes the output held at particular level or was proportional to the input. Over time, responses were evolved that have interesting steps. The fitness function will reward strongly this type of response as there are large differences between the expected and the actual response. Examples of these responses are in figures 8 and 9. It was also noted that the steps disappeared if the analogue signal fed to the LC was slowed (so that it took more than 2ms to apply a voltage sweep), however we have been unable to investigate this phenomena at this time. Figure 11 shows the best individual’s responses from an evolutionary run. The x-axis is voltage in and the y-axis is voltage out. Both axis are in volts. It was noted that small step responses were often generated within only a few generations, typically less than 5 were needed before a noticeable step occurred. To check that the results were found by an evolutionary process and not by random search the previous experiment was repeated but using random search. In this instance, step functions were not frequently observed. Typically no response was obtained from the LCEM. As these results were surprising, it was important to try and discover what was causing the responses observed. The inputs and outputs of the LCEM were verified with an os- cilloscope, and matched with the results that the Evolvatron was recording. The next experiment was to try the evolution without the LCD. The display was removed from the circuit, and the first experiment repeated. No response was found from the EM. This does not demonstrate that the LC itself is responsible, however it removes the chance that it is some feature of the control circuitry. More experiments will be performed in future to try and demonstrate that it is the LC modifying the input signal - however at this stage we have not constructed a suitable ”dummy” display. The next experiment was to try and evolve a transistor. The desired response here is a low output when the input voltage is below a certain threshold and high otherwise. The target threshold was set to 1.5V (in the middle of input voltage range), with a response of 0V below the threshold and at least 3V above. Evolution failed to find a response with the desired characteristics. However, observations of the step functions produced were interesting. It appeared that there were several distinct input voltages at which the steps occurred, and that the switch was from high to low (the oppo- site of the desired transistor). It is not clear what the causes the step to occur at these input voltages and configurations, and it is hoped future experimentation may provide clues to the mechanism. However it is our current opinion that these may be re- lated to the energy thresholds of transitions in state of the LC, where a certain voltage is required before the molecules in the LC will change their orientation. The following experiment describes a more unconstrained approach to evolving a transistor, where no target switch voltage was defined. Following the observation of the step-like responses at particular input voltages, an experiment was set up to try to identify ranges of voltages where evolution could evolve a switching response, i.e. where the LCD exhibits non-linear behaviour to an input voltage. This is essentially the same as attempting to evolve a transistor, however in this experiment there was no requirement for a switch to occur at any particular input voltage. In this experiment the error on each expected value was weighted (by D) according to its distance from the target switch voltage(T). T was varied between 0.1V and 3V at increments of 0.2V. Figure 12 shows the responses of the best individuals from a typical run. Here the target voltage for a step was 2V. We can see that evolution is able to move the solution toward the target but cannot position the step exactly at 2V. From figure 13 we can see that there appear to be step functions that cannot be obtained. The diagram maps areas of activity in the liquid crystal by showing at which voltages the response becomes non-linear(a), and what degree of step change we see at these voltages(b). The horizontal axis shows what the voltage was when a response occurred, the vertical what voltage was read from the LCEM. The darker the spot the more frequently a step was observed at this point. If the response from the liquid crystal was purely linear the map would appear all white, however we see that the liquid crystal is capable of behaving in a non-linear way at large range of voltages. However it’s ability to do this is non-uniform, in other words it is harder to evolve step functions at certain transition voltages than others. The response seen in figure 12 is in a region where a lower number of steps seems obtainable. This is the first time that evolution has been attempted in liquid crystal, and these experiments have demonstrated that a LCD can be used as a FPMA. The experiments show that evolutionary refinement can be used to adjust the response. Although these experiments are very limited, they show that the FPMA model described is feasible and that intrinsic evolution has the potential to exploit the physics of a complex system in a controllable manner. More work is required to prove that the LC is responsible for the observed results and to attempt to discover how the LC is being ex- ploited. This material forms part of a project called DEEPER: Discovery and Exploitation of Emergent Physics through Evolutionary Refinement. It is supported by the Eu- ropean Office of Aerospace Research and Development (EOARD), Airforce Office of Scientific Research, Airforce Research Laboratory, under Contract No. F61775- 02-WE036. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the EOARD. The project is an Emblem project : Further information is available ...
Similar publications
Organizations often create and employ artifacts to intentionally change their routines, but these changes often produce unintended consequences. In this paper, I explore how organizations design artifacts to intentionally change routines by analyzing a series of design performances—organizational actions intended to create an artifact to change a r...
Violation of maxims is a phenomenon where a speaker violates maxims intentionally to produce implicatures. This study discusses the analysis of the flouting maxim using a YouTube video on the BBC Radio 1 YouTube channel "Kids ask Ryan Reynolds". The purpose of this study is to describe the flouting maxim and describe the function of the flouting ma...
Deep Neural Networks are built to generalize outside of training set in mind by using techniques such as regularization, early stopping and dropout. But considerations to make them more resilient to adversarial examples are rarely taken. As deep neural networks become more prevalent in mission-critical and real-time systems, miscreants start to att...
The effects of electrolyte additives on gas evolution, gas consumption and impedance growth at elevated temperature have been studied using Li[Ni0.42Mn0.42Co0.16]O2 (NMC442)/graphite pouch cells and pouch bags containing delithiated NMC442 or lithiated graphite electrodes plus electrolyte. It was found that there was much more gas, mostly CO2, gene...
Several machine learning models, including neural networks, consistently
misclassify adversarial examples---inputs formed by applying small but
intentionally worst-case perturbations to examples from the dataset, such that
the perturbed input results in the model outputting an incorrect answer with
high confidence. Early attempts at explaining this...
Citations
... Julian was also a pioneer in the field of in materio computing (Miller & Downing, 2002), which exploits the physical properties of unconventional materials, such as liquid crystals (Harding & Miller, 2004) and carbon nanotubes (Miller et al., 2014), to perform computation intrinsically, in what he dubbed a "Field Programmable Matter Array." His original work used evolutionary algorithms directly to configure the materials. ...
... To date, work with EiM has explored liquid crystal substrates, composites of randomly dispersed carbon nanotubes mixed with polymers, carbon nanotubes mixed with liquid crystal, and networks of gold nanoparticles (Broersma et al. 2017;Harding and Miller 2004;Miller and Downing 2002). Each of these substrates is evolved to achieve interesting computational properties on a variety of specific tasks, without it being known exactly how best to program them to perform those tasks. ...
We overview Reservoir Computing (RC) with physical systems from an Unconventional Computing (UC) perspective. We discuss challenges present in both fields, including encoding and representation, or how to manipulate and read information; ways to search large and complex configuration spaces of physical systems; and what makes a “good” computing substrate.
... As we have reviewed elsewhere [55], there is a small amount of work considering the evolutionary design of physical systems directly, stretching back to the origins of the discipline [8,51,19,57]. Well-known examples include robot controller design [45]; the evolution of vertebrate tail stiffness in swimming robots [42]; adaptive antenna arrays [4]; electronic circuit design using programmable hardware [66]; product design via human-provided fitness values [28]; chemical systems [65]; unconventional computers [27]; robot embodied evolution [22]; drug discovery [64]; functional genomics [39]; adaptive optics [63]; quantum control [36]; fermentation optimization [18]; and the optimization of analytical instrumentation [46]. A selection of multiobjective case studies can be found in [40]. ...
Design mining is the use of computational intelligence techniques to iteratively search and model the attribute space of physical objects evaluated directly through rapid prototyping to meet given objectives. It enables the exploitation of novel materials and processes without formal models or complex simulation. In this article, we focus upon the coevolutionary nature of the design process when it is decomposed into concurrent sub-design-threads due to the overall complexity of the task. Using an abstract, tunable model of coevolution, we consider strategies to sample subthread designs for whole-system testing and how best to construct and use surrogate models within the coevolutionary scenario. Drawing on our findings, we then describe the effective design of an array of six heterogeneous vertical-axis wind turbines.
... This makes the material cheap and readily available. Liquid Crystal Matrices have successfully been used as a substrate for computation in several experiments [10][11][12]14]. ...
Evolution-in-Materio (EIM) is a method of using artificial evolution to exploit physical properties of materials for computation. It has previously been successfully used to evolve a multitude of different computational devices implemented in physical materials. One of the biggest problems in exploiting materials is finding a good computational abstraction to carry computation on top of the underlying physical process. This paper presents elementary cellular automata (CA) as a possible abstraction and presents successfully evolved CA transition functions in single-walled carbon nanotube (SWCNT) and polymer composite materials. Such implementation allows reasoning about the computational capabilities of materials and draw analogies with cellular automata complexity and computation at the Edge of Chaos. This work is done within the European Project NASCENCE.
... LC contains several key features including, wide availability, addressable using digital voltages, exhibits emergent behaviour, has a unique mesomorphic structure between ordered and disordered, and can relax to an initial base state. Harding & Miller [39] adopt liquid crystal as a basis material and construct a bespoke platform to solve multiple computational problems. The hardware houses a liquid crystal display (LCD) and an array of dynamically selectable input/output connections to both the LCD and external measurement devices. ...
... They demonstrate liquid crystal as an efficient evolvable material where relatively small numbers of generations can produce effective solutions. Over the course of their investigation the LC system has been applied to three separate tasks; tone discrimination [39], creating logic gates [41] and a real-time robot controller [40]. ...
Research in substrate-based computing has shown that materials contain rich properties that can be exploited to solve computational problems. One such technique known as Evolution-in-materio uses evolutionary algorithms to manipulate material substrates for computation. However, in general, modelling the computational processes occurring in such systems is a difficult task and understanding what part of the embodied system is doing the computation is still fairly ill-defined. This chapter discusses the prospects of using Reservoir Computing as a model for in-materio computing, introducing new training techniques (taken from Reservoir Computing) that could overcome training difficulties found in the current Evolution-in-Materio technique.
... For example, the first ESs were used to design jet nozzles with a string of section diameters, which were then machined and tested for fitness (Rechenberg, 1971). Other well-known examples include robot controller design (Nolfi, 1992), electronic circuit design using programmable hardware (Thompson, 1998), product design via human provided fitness values (Herdy, 1996), chemical systems (Theis et al., 2006), and unconventional computers (Harding and Miller, 2004). More recently, Boria et al. (2009) used an EA to evolve a morphing wing structure where physical designs were morphed using a set of actuators and evaluated in a closed-loop wind tunnel. ...
An initial study of surrogate-assisted evolutionary algorithms used to design
vertical-axis wind turbines wherein candidate prototypes are evaluated under
approximated wind tunnel conditions after being physically instantiated by a
three-dimensional printer has recently been presented. Unlike other approaches,
such as computational fluid dynamics simulations, no mathematical formulations
were used and no model assumptions were made. This paper extends that work by
exploring alternative surrogate modelling and evolutionary techniques. The
accuracy of various modelling algorithms used to estimate the fitness of
evaluated individuals from the initial experiments is compared. The effect of
temporally windowing surrogate model training samples is explored. A
surrogate-assisted approach based on an enhanced local search is introduced;
and alternative coevolution collaboration schemes are examined.
... The evaluation of physical artifacts for fitness determination can be traced back to the origins of evolutionary computation; for example, the first evolution strategies (ESs) were used to design jet nozzles with a string of section diameters, which were then machined and tested for fitness [13]. Other wellknown examples include robot controller design [14], electronic circuit design using programmable hardware [15], product design via human provided fitness values [16], chemical systems [17], and unconventional computers [18]. Evolution in hardware has the potential to benefit from access to a richer environment where it can exploit subtle interactions that can be utilised in unexpected ways. ...
The production of renewable and sustainable energy is one of the most important challenges
currently facing mankind. Wind has made an increasing contribution to the
world's energy supply mix, but still remains a long way from reaching its full
potential. In this paper, we investigate the use of artificial evolution to
design vertical-axis wind turbine prototypes that are physically instantiated
and evaluated under approximated wind tunnel conditions. An artificial neural
network is used as a surrogate model to assist learning and found to reduce the
number of fabrications required to reach a higher aerodynamic efficiency.
Unlike in other approaches, such as computational fluid dynamics simulations,
no mathematical formulations are used and no model assumptions are made.
Initially a conventional evolutionary algorithm is used to explore the design
space of a single wind turbine and later a cooperative coevolutionary algorithm
is used to explore the design space of an array of wind turbines.
... For example, the first evolution strategies were used to design jet nozzles with a string of section diameters, which were then machined and tested for fitness [63]. Other well-known examples include robot controller design [54], electronic circuit design using programmable hardware [77], product design via human provided fitness values [33], chemical systems [76], and unconventional computers [30]. More recently, [9] used an EA to evolve a morphing wing structure where physical designs were morphed using a set of actuators and evaluated in a closed-loop wind tunnel. ...
We have recently presented an initial study of evolutionary algorithms used to design vertical-axis wind turbines (VAWTs) wherein candidate prototypes are evaluated under fan generated wind conditions after being physically instantiated by a 3D printer. That is, unlike other approaches such as computational fluid dynamics simulations, no mathematical formulations are used and no model assumptions are made. However, the representation used significantly restricted the range of morphologies explored. In this paper, we present initial explorations into the use of a simple generative encoding, known as Gielis superformula, that produces a highly flexible 3D shape representation to design VAWT. First, the target-based evolution of 3D artefacts is investigated and subsequently initial design experiments are performed wherein each VAWT candidate is physically instantiated and evaluated under fan generated wind conditions. It is shown possible to produce very closely matching designs of a number of 3D objects through the evolution of supershapes produced by Gielis superformula. Moreover, it is shown possible to use artificial physical evolution to identify novel and increasingly efficient supershape VAWT designs.
... Different types of material have the potential to be used within this concept. Liquid crystals have been used to evolve a tone discriminator [7,11], a real-time robot controller [8,9] and logic gates [10]. Randomly generated stochastic networked structures formed inside a piece of material have been used for developing memristor networks with neuron-like behavior. ...
... The AND and OR gate are easy to resolve, since it takes just one configuration voltage to achieve that even when i.e. the same voltage is used to denote a logic 1 in the input and the output in view of equation (11). For the XOR gate, the following thresholds are used. ...
This paper presents results of computations based on threshold logic performed by a thin solid film, following the general principle of evolution in materio. The electrical conductivity is used as the physical property manipulated for evolving Boolean functions. The material used consists of a composite of single-wall carbon nanotubes (SWCNTs) and the polymer poly(methyl methacrylate). The SWCNTs are randomly dispersed in the polymer forming a complex conductive network at the nano-scale. The training is formulated as an optimisation problem with continuous and binary constraints and is subsequently solved by two derivative-free algorithms, the Nelder-Mead (NM) and the Differential Evolution (DE) algorithms. This approach has been used to evolve gates and circuits. The NM fails to converge for all computational tasks, whereas the DE is always successful. The computation tasks considered are simple threshold logic gates and more complicated circuits. The thin film composite is very stable and its behavior remains the same after the optimal solution has been achieved.
... Embodied evolutionary computing has typically referred to the existence of a physical solution in the fitness evaluation, and can be traced back to the origins of the discipline: the first evolution strategies (ESs) were used to design jet nozzles as a string of real-valued diameters, which were then machined and tested for fitness [12]. Other well-known examples include robot controller design (e.g., [13]), electronic circuit design using programmable hardware (e.g., [14]), product design via human provided fitness values (e.g., [15]), chemical systems (e.g., [16]), and unconventional computers (e.g., [17] ). Evolution in hardware has the potential to benefit from access to a richer environment where it can exploit subtle interactions that can be utilised in unexpected ways. ...
Renewable and sustainable energy is one of the most important challenges
currently facing mankind. Wind has made an increasing contribution to the
world's energy supply mix, but still remains a long way from reaching its full
potential. In this paper, we investigate the use of artificial evolution to
design vertical-axis wind turbine prototypes that are physically instantiated
and evaluated under approximated wind tunnel conditions. An artificial neural
network is used as a surrogate model to assist learning and found to reduce the
number of fabrications required to reach a higher aerodynamic efficiency,
resulting in an important cost reduction. Unlike in other approaches, such as
computational fluid dynamics simulations, no mathematical formulations are used
and no model assumptions are made.
