FEA and Darwinian Computer Simulations

_{Gil Dodgen

June 11, 2011

Intelligent Design}

Share: Facebook; Twitter; LinkedIn; Flipboard; Print; Email

In my work as a software engineer in aerospace R&D I use what is arguably the most sophisticated, universally applicable, finite-element analysis program ever devised by the most brilliant people in field, refined and tested for 35 years since its inception in the mid-1970s for the development of variable-yield nuclear weapons at Lawrence Livermore National Laboratory. It is called LS-DYNA (LS for Livermore Software, and DYNA for the evaluation of dynamic, nonlinear, transient systems).

A finite element is an attempt to descretize on a macro level what occurs at a molecular level in a physical system, so that a result is amenable to a practical computational solution. The learning curve for the use of this sophisticated technology is extremely steep, and the most important thing one learns is that empirical verification of the simulation results is absolutely required to validate the predictions of any FEA model.

In an LS-DYNA simulation, all the laws of physics and the mathematics that describe them are precisely known. In addition, all of the material properties associated with the physical objects are precisely quantified with empirical verification (density, modulus of elasticity, and much more).

The FEA solver computes a physical result by solving millions of differential equations with a minimal integration time step based on the time required for a disturbance traveling at the speed of sound to traverse the smallest finite element with the greatest mass density.

Even with all of this, and countless man-years of experience by sophisticated and experienced users (LS-DYNA has been used for many years in the auto industry for simulating car crashes) empirical verification is always required, by actually crashing a car to validate the FEA results.

In light of all this, consider the typical Darwinian computer simulation and the trust that could be put in one.

Darwinian computer simulations are simply a pathetic joke as they relate to biological reality. This should be obvious to anyone with experience in the field of legitimate computer simulation.

Comments

Charles:
That has about as much relevance to this discussion as studying computer viruses for antibiotic research. There is nothing “genetic” about circuit design algorithms except in the fevered imaginations of engineer wannabes. Iterative formulaic revision is not evolution.
We were discussing Genetic Algorithms, so I guess an example of a genetic algorithm in use is irrelevant, yes ;) Its not biological evolution, it is artificial evolution. Both processes involve replication with variance and differential rates of replication. Both processes are evolution, but only one is biological. I'm not an 'engineer wannabe' I'm just an engineer and scientist. My wife tells me I'm quite a talented chef as well :)
I daresay there exists within your respective disciplines and literature an unwitting collective self-congratulatory exaggeration about the state of the art of evolutionary theory modelling.
I'm not discussing evolutionary theory modelling in the main, I'm discussing how Genetic Algorithms work from an engineering perspective although I have tried to put it in some context with respect to biology.
If you want to understand the poor state of evolutionary theory modelling, relative to engineering and the hard sciences, start with the sloppy conceptualizations used to obscure a lack of detailed factual understanding.
Lack of detailed understanding of what? How to use GA's to solve design problems?
if you say they have “evolvable genetic algorithims” instead of being reconfigured or reprogrammed you will likely lose credibility.
Well for my stuff certainly because it is irrelevant, although you can use GA's to optimise motion control algorithms. There are plenty of companies out there that do use genetic algorithms to create stuff, and they get venture capital and make a lot of money. Try these guys at Naturalmotion. They design software for computer games and animation studios that lets you use a virtual stuntman or character, equipped with virtual muscles and reflexes modelled on biology. They use, amongst other things, genetic algorithms to craft behaviours. Venture capitalists will give money for something that uses a GA if it is demonstrated that they could make money from it - i.e. that it works.DrBot_{June 16, 2011
June
06
Jun
16
16
2011
01:48 PM
1
01
48
PM
PDT}

Mung:
The population size is also an aspect of the design of a GA.
Yes, but not always. You can design the population size to be variable if you choose, although you run into practical problems of computer power and memory requirements. For engineering purposes there doesn't seem much point it varying the population (but I may be wrong) It is only relevant if studying biology, and even then it isn't always relevant to the particular aspect of evolution being studied. If you are studying population dynamics then I guess variable populations in your model is fairly critical :)DrBot_{June 16, 2011
June
06
Jun
16
16
2011
01:28 PM
1
01
28
PM
PDT}

DrBot:
Mung, here is an example from engineering of how to use a GA to evolve an electronic circuit
Well yeah, BY DESIGN! And we do not understand biological organisms well enough to simulate their evolution.Joseph_{June 16, 2011
June
06
Jun
16
16
2011
01:28 PM
1
01
28
PM
PDT}

Damn, my last comment has disappeared under a discussion, and it's still in moderation. Gil - I asked something at 48, if/when you see this, I'd be interested to see your response.Heinrich_{June 16, 2011
June
06
Jun
16
16
2011
01:25 PM
1
01
25
PM
PDT}

Mung:
That’s what I have been arguing. I fail to see why the question is even in dispute.
I've already explaind but I'll try and expand. The first problem (partly due to wikipedia) is that the terms used are hijacked from biology and applied to evolutionary computing, and different people use different terms. Within the circles I work it it is fairly normal to refer to the candidate solution as the genome, not the chromosome - but others, as you illustrate from wikipedia, use different terminology. In the end they are just words, we are both talking about the same things but with different terminology and it is the thing, not the word that is under discussion. I'm happy to use chromosome of you prefer?
The mutation operator and crossover operator employed by the genetic algorithm must take into account the chromosome’s design.
Quite true but that doesn't mean that the chromosome has to have been designed - thats just bad terminology I suspect ;) - as I already said, when you use a GA on some systems the chromosome is intrinsic to the system so you don't need to design it, but you do need to understand it in order to have mutation and crossover that actually works with the particular encoding. It is certainly true though that many scientists that use GA's for design tasks very carefully design encoding schemes and there is a lot of published work on this area. When it comes to modeling biology though what we would really want is to directly simulate biology, so we are designing the simulator to replicate the operation of a biological encoding. we are not designing the encoding scheme ourselves (even though it might have been designed by something else)DrBot_{June 16, 2011
June
06
Jun
16
16
2011
01:11 PM
1
01
11
PM
PDT}

The population size is also an aspect of the design of a GA.Mung_{June 16, 2011
June
06
Jun
16
16
2011
12:57 PM
12
12
57
PM
PDT}

DrBot @70: (GA’s have been used successfully to evolve circuits on FPGA’s) That has about as much relevance to this discussion as studying computer viruses for antibiotic research. There is nothing "genetic" about circuit design algorithms except in the fevered imaginations of engineer wannabes. Iterative formulaic revision is not evolution. Elizabeth Liddle @71: Another thing I do is “mate” solutions – I draw individuals at random from my population, the fitter ones having greater probability of being drawn, and then “mate” them with a second randomly drawn individual, and randomly recombine the genomes. That speeds things up a bit, and widens the search space, by loosening the linkage between useful and not-useful bits of the genome. As in life And if a Darwinism/Evolution critic were to claim that in life, individuals mate at random you would instead be lecturing us on phenotypic frequencies and adaptive genotypic selection pressures, anything but "random" mating behavior. If the researchers engineering new corn, wheat, soybean hybrids followed your lead, the world would starve. Almost nothing about life or the real world is random. Even brownian motion and weather are not random but predictible when sufficient measurements and observations are known. The two of you both are carelessly tossing double-entendre euphemisms around, as if enclosing them in quotes compensates for their fundamental misapplication. While you both consider yourselves aware of the limits to which you apply your metaphors and euphemisms, I daresay there exists within your respective disciplines and literature an unwitting collective self-congratulatory exaggeration about the state of the art of evolutionary theory modelling. Consider where you're now at versus Gil's original point:
Darwinian computer simulations are simply a pathetic joke as they relate to biological reality. This should be obvious to anyone with experience in the field of legitimate computer simulation.
If you want to understand the poor state of evolutionary theory modeling, relative to engineering and the hard sciences, start with the sloppy conceptualizations used to obscure a lack of detailed factual understanding. Elizabeth, you correctly noted (on a different discussion thread) the importance of a correct and detailed problem statement or restatement to its solution. I submit that the problem of simulating evolutionary theory without adhering to the precise meaning of terms like "in life" (i.e., in vivo), "evolution" (which is not programmed, iterative non-self-repliating revision), "random mutation" and "natural selection" will lead to innumerable irrelevant results. Conversely, DrBot, I submit that were you to seek funds from venture capitalists for your robotic servo startup, and one of them questioned how adaptable they were to changing robotic designs, if you say they have "evolvable genetic algorithims" instead of being reconfigured or reprogrammed you will likely lose credibility. Investors in companies want to understand how a product actually works, not be snowed with exaggerated marketing language.Charles_{June 16, 2011
June
06
Jun
16
16
2011
12:51 PM
12
12
51
PM
PDT}

Chromosome Design
I think Genome is more appropriate than chromosome...But if you prefer chromosome then that's fine.
Well, if it's a wiki war you want! ;)
In genetic algorithms, a chromosome (also sometimes called a genome) is a set of parameters which define a proposed solution to the problem that the genetic algorithm is trying to solve. The chromosome is often represented as a simple string, although a wide variety of other data structures are also used. - Chromosome (genetic algorithm)
I just love the next section on that page. It's called Chromosome design
The design of the chromosome and its parameters is by necessity specific to the problem to be solved. The mutation operator and crossover operator employed by the genetic algorithm must take into account the chromosome's design.
That's what I have been arguing. I fail to see why the question is even in dispute.Mung_{June 16, 2011
June
06
Jun
16
16
2011
12:50 PM
12
12
50
PM
PDT}

First, you have to decide that it’s actually going to be a bit string, that you’re going to represent candidate solutions using the two bits 0 and 1. That is a design decision. Second, you have to decide what each series of bits in the chromosome means. That is a design decision.
Not for the person evolving the circuit in my example. The FPGA is designed, obviously, and the method of configuring it is part of the design but it was not designed as a system for evolving circuits. In fact until Thompsons work most people didn't think you could evolve them, or it hadn't occurred to them to try. When it comes to evolving them - applying a GA - the encoding scheme is not designed by the person using the GA, it is an intrinsic part of the system being evolved so from that perspective it is imposed by the system and not a design choice. Of course you can design an indirect encoding like the one I described, in which case yes, you are designing the encoding scheme, but I still think it is incorrect to say you are designing the genome or chromosome. We may just be quibbling over semantic details here - when I say 'designing a genome' I mean manually configuring the contents of the genome, not specifying the parameters within which a genome can operate, and how it is interpreted or mapped to a phenotype.
Getting the chromosome right is a huge part, one might even say a necessary part, of getting the GA to solve the problem.
As I've pointed out above, sometimes you don't need to design it because it is intrinsic to the thing you are applying the GA to - the GA is just manipulating the configuration of a system. Now if we map this back to biology, which is what I suspect you are trying to make a point about. It makes no difference if the method of encoding and inheriting information in a cell or organism is the product of deliberate design or not. If first life was designed, rather than a product of some complex chemistry, then the encoding scheme was designed, otherwise it was the product of natural forces. Either way Evolution still occurs - Genomes get replicated with variance, resulting in differential survival rates. Evolution is a process that occurs when you have replication with variance resulting in differential rates of subsequent replication. The origin of the replicator (design or chemistry) makes no difference. God can design life to evolve, and we can study and replicate the processes.DrBot_{June 16, 2011
June
06
Jun
16
16
2011
12:47 PM
12
12
47
PM
PDT}

Mung, I think Genome is more appropriate than chromosome. From wikipedia
In modern molecular biology and genetics, the genome is the entirety of an organism's hereditary information.
Although none of the terms are entirely 'right' - the FPGA is 'inherited' in one sense but in another you could say that it is an environment and every individual gets a 'turn' in the environment. Perhaps it is best to just consider the elements of inheritance that are subject to variation (the bit string) as being like a genome and the FPGA as being like the cell body, but again these are analogies to a term derived from biology. It only makes sense to take them literally if we are actually talking about GA's for direct modeling of biology, not as design tools for engineers. But if you prefer chromosome then thats fine.DrBot_{June 16, 2011
June
06
Jun
16
16
2011
12:23 PM
12
12
23
PM
PDT}

Search Space If you have a bit string of length 3, you will have 8 possible configurations of the bit string. 3^2 = 8 That's the size of the search space. The only way to change increase the size is to increase the length of the bit string. You don't increase the size by swapping bits around, which is typically all that happens during crossover.Mung_{June 16, 2011
June
06
Jun
16
16
2011
12:16 PM
12
12
16
PM
PDT}

Search Space Elizabeth Liddle:
Either way, crossover tends to widen the search space because without crossover, linkage tends to restrict the searchable space.
You're not using "search space" in the way I mean the term. And I don't think you're using it according to the accepted meaning of the term. See my post @62. Also:
If we are solving some problem, we are usually looking for some solution, which will be the best among others. The space of all feasible solutions (it means objects among those the desired solution is) is called search space (also state space). Each point in the search space represent one feasible solution. Each feasible solution can be "marked" by its value or fitness for the problem. We are looking for our solution, which is one point (or more) among feasible solutions - that is one point in the search space. The looking for a solution is then equal to a looking for some extreme (minimum or maximum) in the search space. The search space can be whole known by the time of solving a problem, but usually we know only a few points from it and we are generating other points as the process of finding solution continues. - Search Space
So with mutation and crossover you are generating new points in the search space, not changing the size of the search space itself.
Search Space - All possible solutions to the problem - Genetic Algorithms Overview
Changing a chromosome does not equate to increasing the number of possible solutions.Mung_{June 16, 2011
June
06
Jun
16
16
2011
12:07 PM
12
12
07
PM
PDT}

Crossover Elizabeth Liddle:
Crossover can change the length of the bit string. I’ve done it both ways.
It can. Only one version of crossover listed on Wikipedia changes the length of the strings, "Cut and Splice." http://en.wikipedia.org/wiki/Crossover_%28genetic_algorithm%29 But if it does lead to a child that has a longer or shorter bit string, that has to be taken into account and is a part of the design.Mung_{June 16, 2011
June
06
Jun
16
16
2011
11:54 AM
11
11
54
AM
PDT}

How are you defining each, in this context, or are you using them interchangeably?
I've been using them interchangeably. I'd prefer to use chromosome. See my post @62.Mung_{June 16, 2011
June
06
Jun
16
16
2011
11:40 AM
11
11
40
AM
PDT}

Elizabeth, Point taken, we may be getting caught up in terminology - in the case of the FPGA the search space is defined by the permitted length of the bit string, which could be the total configuration possibilities for the piece of hardware, or it could be deliberately limited. This is different to the searchable space, which may only be a fraction of the actual space available and, as you point out, is determined partly by the particular GA used (and of course the topology of the search space)DrBot_{June 16, 2011
June
06
Jun
16
16
2011
11:37 AM
11
11
37
AM
PDT}

DrBot:
So to answer your question, assuming I understood it, the genome or chromosome is not designed (apart from by the GA) but the encoding scheme can either be designed, chosen or imposed, depending on the task.
The chromosome is not designed by the GA. It is at times modified by one or or more operators, such as mutation, crossover, and selection. Those operators are designed. There is no "willy-nilly" about it. The encoding scheme has to be encoded in the chromosome. That's what I mean by "the chromosome is designed." Let's take your basic bit string as an example. First, you have to decide that it's actually going to be a bit string, that you're going to represent candidate solutions using the two bits 0 and 1. That is a design decision. Second, you have to decide what each series of bits in the chromosome means. That is a design decision. Even the choice to start the chromosomes by seeding them with a random sequence of bits is a design decision. Why not start them with all 0's, or all 1's? There is a coherence between the bit string and the problem that you're trying to solve. That coherence is designed. If it were not you would have little to no hope of coming up with a solution to your problem by modifying the candidate solutions. Getting the chromosome right is a huge part, one might even say a necessary part, of getting the GA to solve the problem. True? Now while neither you nor Elizabeth has come right out and stated that the chromosome does not need to be designed, you certainly leave room for people to think you are saying so. So if you want to continue in that vein, please come right out and state that how the chromosome is configured doesn't matter. Then we can put that claim to the test. But we all know that it does matter. :)Mung_{June 16, 2011
June
06
Jun
16
16
2011
11:35 AM
11
11
35
AM
PDT}

Mung: you have used the words "genome" and "chromosome" with regard to a GA. How are you defining each, in this context, or are you using them interchangeably?Elizabeth Liddle_{June 16, 2011
June
06
Jun
16
16
2011
11:17 AM
11
11
17
AM
PDT}

Crossover can change the length of the bit string. I've done it both ways. If you randomise the cut points, you can, as in life, end up with a shorter or longer string. Either way, crossover tends to widen the search space because without crossover, linkage tends to restrict the searchable space.Elizabeth Liddle_{June 16, 2011
June
06
Jun
16
16
2011
11:15 AM
11
11
15
AM
PDT}

Elizabeth: "Another thing I do is “mate” solutions ... That speeds things up a bit, and widens the search space..." ME: "I think this language is imprecise. You’re not widening the size of the search space." DrBot: "I think that's correct for the example I gave – the size of the search space is determined by the length of the bit string." I think you're agreeing with me, lol. I just want to make sure. Elizabeth was talking of using crossover which she referred to as mating. Crossover doesn't increase the length of the bit string.
...the size of the search space is determined by the length of the bit string.
I agree. Crossover ("mating") does not widen the search space. I just felt that she probably meant to say it "widens" the search in some way, not that it changes the size of the search space.Mung_{June 16, 2011
June
06
Jun
16
16
2011
11:10 AM
11
11
10
AM
PDT}

Mung
I think this language is imprecise. You’re not widening the size of the search space.
I think thats correct for the example I gave - the size of the search space is determined by the length of the bit string. It is possible to use variable length genomes though, but in the case of the example I gave the total search space would still be the length of a bit string required to fully configure an FPGA, but candidate solutions could have a genotype much shorter than this (and consequently only use a limited portion of the FPGA real estate. You can, in theory, use a variable length genome with no length limit, rendering the search space infinite, but this isn't very useful (and on a computer you have an effective limit imposed by memory). The example I gave is one of direct genotype to phenotype mapping. You could extend the experiment to an indirect mapping (although I don't know if it would help things) Basically your genotype could encode an FPGA configuration, and you then use this configuration to 'clock out' a new bit string that encodes for a new FPGA configuration which is then used to test the design fitness. Like I said, I have no idea if this would be useful but some indirect encodings can be powerful, for example by coding for repeating patterns so the pattern is only described once in the genotype, but gets repeated many times for the phenotype. To your earlier post,
That’s a bit ambiguous. It’s the fpga that is the phenotype in the example, correct? Not the binary string?
Correct.
So all that remains, at this point, imo, is to discuss whether the genotype in his example is designed.
I suppose you could describe it that way but it seems a little odd. The encoding scheme is chosen but the genotypes of individuals are random (to start with) within the bounds of the encoding scheme. In the case of the FPGA the encoding scheme is not really designed or chosen, it is imposed by the hardware - these chips are designed to be configured by a bit string so we are just using a GA to generate a bit string that will produce behavior we want. So to answer your question, assuming I understood it, the genome or chromosome is not designed (apart from by the GA) but the encoding scheme can either be designed, chosen or imposed, depending on the task.DrBot_{June 16, 2011
June
06
Jun
16
16
2011
09:31 AM
9
09
31
AM
PDT}

GA’s have been used successfully to evolve circuits on FPGA’s
Not an issue in dispute. What is in disputes is how analogous is a GA to biological evolution. What aspects of a GA are designed, and which aspects are not. The more aspects of a GA that are designed, and the closer the analogy to biological evolution, the stronger the case that biological evolution is designed. Lizzie was only allowing for design at two points in a GA, I claimed there were more, and that what was being left out was a significant aspect of the function of a GA. e.g., the chromosome itself.Mung_{June 16, 2011
June
06
Jun
16
16
2011
09:10 AM
9
09
10
AM
PDT}

Elizabeth:
Another thing I do is “mate” solutions ... That speeds things up a bit, and widens the search space,
I think this language is imprecise. You're not widening the size of the search space.Mung_{June 16, 2011
June
06
Jun
16
16
2011
09:00 AM
9
09
00
AM
PDT}

Elizabeth Liddle @69:
Well, looks like we had better use some specific examples. Would you like to pick a GA that you have in mind?
First and foremost, let's be clear that at this time I am talking about one thing and one thing only. The "chromosome" or "genotype" in a GA. I have three claims: 1. You left the chromosome out of your list. 2. The chromosome itself is designed. 3. That's a significant oversight.
Well, looks like we had better use some specific examples. Would you like to pick a GA that you have in mind?
Take DrBot's example @70:
You have an FPGA (reconfigurable array of logic gates). The configuration is determined by a binary string (lots of 1?s and 0?s). This is the phenotype. It is connected to some test equipment (the environment).
That's a bit ambiguous. It's the fpga that is the phenotype in the example, correct? Not the binary string?
Lets generate a starting population of 30 – we generate 30 entirely random bit strings (genotypes).
In the terms I've been using, those bit strings (binary strings) is the chromosomes (or genotypes). And that's what I am talking about. So far, I think DrBot and I are on the same page. So all that remains, at this point, imo, is to discuss whether the genotype in his example is designed. I say the chromosome is designed. It has to work with the other aspects of the GA, such as the phenotype and fitness function. It's not just some isolated non-designed entity that just happens to accidentally function in the context of the GA. But do we really need to debate that? Don't you both already know it is the case?
...load the population member (random bitstring) into the fpga (map the genotype to the phenotype) then perform a fitnes test...
I mentioned the mapping requirement in an earlier post. Mung @61
2. There needs to be a mapping (also designed) of the individuals to the designed fitness function.
Now if DrBot's GA is not enough, if you want a different example, I have two I can suggest. The first is the ev program. The Java source code is available online. The second is we could look at the GA at: http://www.cleveralgorithms.com/ The entire book and code is online.Mung_{June 16, 2011
June
06
Jun
16
16
2011
08:50 AM
8
08
50
AM
PDT}

heh. I did my PhD in a motor control lab, and I agree :)Elizabeth Liddle_{June 16, 2011
June
06
Jun
16
16
2011
06:39 AM
6
06
39
AM
PDT}

Apart from a bit of undergraduate teaching (non symbolic AI) I've just set up a company, currently in stealth mode, developing electric servo actuator systems for autonomous robots where the servos can be programmed to behave as compliant mechanisms (for example like antagonistic muscles) rather than the normal rigid systems used in most robots today. I'm firmly of the opinion that most robots built today are somewhat of a dead end because they treat actuators and the joints they control as sources of movement only - they impose motion - whereas biological joints are variable stiffness mechanisms that can accept kinematic inputs from the environment, as well as generating motion. A really good example would be passive dynamic and ballistic walking, which Cornell university has done some seminal work on. When I say I just set up the company it basically means I'm working from home with almost no income and a 2yo daughter to manage ;) the UK job market is a bit sparse at the moment, even in the sciences!DrBot_{June 16, 2011
June
06
Jun
16
16
2011
04:55 AM
4
04
55
AM
PDT}

Cool! What do you work on now?Elizabeth Liddle_{June 16, 2011
June
06
Jun
16
16
2011
04:33 AM
4
04
33
AM
PDT}

You can find a page of early work by Adrian Thompson on evolvable hardware here (Adrian was one of the people who examined my PhD thesis)DrBot_{June 16, 2011
June
06
Jun
16
16
2011
04:21 AM
4
04
21
AM
PDT}

Another thing I do is "mate" solutions - I draw individuals at random from my population, the fitter ones having greater probability of being drawn, and then "mate" them with a second randomly drawn individual, and randomly recombine the genomes. That speeds things up a bit, and widens the search space, by loosening the linkage between useful and not-useful bits of the genome. As in life :)Elizabeth Liddle_{June 16, 2011
June
06
Jun
16
16
2011
04:12 AM
4
04
12
AM
PDT}

Mung, here is an example from engineering of how to use a GA to evolve an electronic circuit The experimental setup: You have an FPGA (reconfigurable array of logic gates). The configuration is determined by a binary string (lots of 1's and 0's). This is the phenotype. It is connected to some test equipment (the environment). The goal: Evolve an 8 bit adder. The process: Lets generate a starting population of 30 - we generate 30 entirely random bit strings (genotypes). Now we need a method of evaluating fitness - load the population member (random bitstring) into the fpga (map the genotype to the phenotype) then perform a fitnes test - generate two 8 bit values, apply them to the FPGA inputs, clock the system and measure 8 outputs, score the result by seeing how close the output is to what addding these two 8 bit values ought to be. Repeat this a few times with different random inputs and average the results to get the final fitness score. Now for reproduction and mutation (using the simplest, but not most effective method): Pick two individuals at random, test the fitness of each using the method described above. Overwrite the less fit individual with the fitter one (reproduction), then pick a random number of bits in the bitstring of the individual that was copied and flip them (Mutation). Now repeat this process a few thousand times, or impliment an algorithm that will terminate the process if the average fitness stops increasing for too long (the population gets stuck on a local maxima) or fitness reaches a high enough score. Basically individuals that score higher have a greater chance of reproducing. What you ought to end up with is an population of FPGA configurations that are reasonably good (but probably not perfect) at adding 8 bit numbers. The results will probably be variable - If you do the experiment several times then some runs would produce much better candidate solutions than others - it depends a lot on the starting populations. (GA's have been used successfully to evolve circuits on FPGA's)DrBot_{June 16, 2011
June
06
Jun
16
16
2011
03:51 AM
3
03
51
AM
PDT}

Well, looks like we had better use some specific examples. Would you like to pick a GA that you have in mind?Elizabeth Liddle_{June 16, 2011
June
06
Jun
16
16
2011
12:32 AM
12
12
32
AM
PDT}

Prev 1 2 3 4 5 Next

You must be logged in to post a comment.

Leave a Reply