S. J. Louis and G. J. Rawlins. Designer Genetic Algorithms: Genetic Algorithms in Structure Design. In Proceedings of the 4th International Conference on Genetic Algorithms, pages 53--60, San Diego California, . 1991. Morgan Kaufmann Publishiers, Inc.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....I Behavioural Description i HDL [ 22, 40] Eqtns. in state space [ 30] I RTL Dataflow I FSM Logic [75, 57] Netlist Device Specific Netlist Function Gates Logic Gates Asynchronous Logic Analog Components Function Level Units Configuration Level Units [23, 52] 36, 48, 67, 68, 74] [8, 27,30, 32, 35, 39, 46, 47, 51, 59, 60, 61, 55, 62, 88, 89, 90, 92,93,94] [80, 82] 2, 18, 41, 49, 50, 79, 99] 28, 11, 66, 99] 21, 43, 44, 46, 78, 80, 81, 82, 83, 84, 85, 86, 87, LESS [ Hardware Behavior I Fig. 3: Levels of abstraction needed to characterize behavior of evolved circuits. Torresen [89] partitioned levels of abstraction into the digital and ....

.... one evolved at a logic netlist level and evaluated extrinsically, and one evolved at a device specific netlist level, evaluated intrinsically. The problem selected for both was the evolution of a two bit adder. This problem has been well studied in the past, initially by Louis and Rawlins [51] and more recently by Miller, Thomson and Fogarty [62] Coello and Aguirre [7] and Hollingworth and Tyrell [29] 4.1 Case Study 1 Logic Level Evolution 4.1.1 Phenotype Abstraction In this experiment we evolve the adder at a reasonably high level of abstraction. The representation is ....

Louis S.J. and Rawlins G.J.E. (1991), Designer Genetic Algorithms: Genetic Algorithms in Structure Design, Proc. of the 4th Int. Conf. on Genetic Algorithms, San Diego, CA, U.S.A., pp. 53-60. 42

....to the LUTs i.e. a threshold value rather than a competition. In this way the cell can be mapped to the Virtex architecture at any point during its development, using the Xilinx JBits API. 3. 5 Experiments Evolving Two Bit Adders The evolution of adders has been well studied in the past [32 35] and the adder design space is very well understood, particularly in the digital domain. We have suggested that one of the primary uses of development is the learning of a bias that embodies a useful design abstraction. It is known that this problem can be described using such a design abstraction ....

S. J. Louis and G. J. E. Rawlins, "Designer Genetic Algorithms: Genetic Algorithms in Structure Design," Proc. of the 4th Int. Conf. on Genetic Algorithms, San Diego, CA, U.S.A, 1991.

....Li, Wei, 346] Lienig, Jens, 226, 265, 366] Lin, Chyi Yeu, 125, 126] Lin, C. S. 262] Lin, C. Y. 10] Lin, Shyh Chang, 55] Lin, Y. L. 110, 111] Lin, Youn Long, 280] Liu, B. D. 202] Liu, Luoping, 215] Liu, Xingzhao, 238, 186, 187, 449] Lohbeck, T. K. 431] Louis, Sushil John, [144, 145, 146] Lucasius, C. B. 365] Lucasius, Carlos B. 208, 147] Luchian, Henri, 17] Lunn, Ken, 367] Luong, L. H. S. 330] Ly, Tai A. 193] M, Russo, 311] Ma, Jianhua, 361] Maeda, K. 317] Magele, C. A. 120, 121, 149] Maher, M. L. 150] Maher, Mary Lou, 20, 34] Mahotilo, K. V. 373] ....

....Yahya, 84] Rajeev, S. 175] Rajroop, P. 225] Raman, S. 295] RamBabu, P. 31] Ranjithan, S. 174] Rao, B. B. Prahlada, 57, 91, 395] Rao, Harish A. 336] Rao, Singiresu S. 178] Rao, Vasant B. 180, 443, 181] Rastogi, M. 316] Ravichandran, B. 183] Rawlins, Gregory J. E. [144, 145] Rebaudengo, M. 235, 260] Reddy, S. M. 240, 448] Reeves, Colin R. 236, 268, 299, 441] Reinartz, Karl Dieter, 444] Renhart, W. 149] Renner, G. 256] Reorda, Matteo Sonza, 235] Reynolds, Robert G. 270] Richards, Dana S. 104, 105] Richards, Gill G. 114] Richards, R. A. 494] ....

[Article contains additional citation context not shown here]

Sushil John Louis and Gregory J. E. Rawlins. Designer genetic algorithms: Genetic algorithms in structure design. In Belew and Booker [510], pages 53--60. ga:Louis91b.

....allow operation under reduced power supply voltages [1] Although, the technique currently operates at gate level its main advantages become apparent when used within a high level framework. Our results are illustrated with examples at the gate level by using the adder and parity checker problems [2] The second technique operates on a Data Flow Graph (DFG) consisting of high level blocks such as registers, adders, and multipliers. Operations such as retiming and automatic pipelining are used to reduce the critical path of the DFG, hence allowing operation under reduced supply voltage. Our ....

....by Cadence, Mentor, Plessy, and ES2) This provides the added flexibility of integration with such tools, hence providing a valuable tool to the VLSI circuit designer who is facing increasingly challenging tasks such as designing for low power, area, speed, etc. In addition, the genetic operators [2] are specifically designed to take the circuit aspects of the structural design into consideration. Our representation and the specifically developed genetic operators are crucial to the flexibility of our genetic algorithm and its superiority over previous implementations. The GA utilises a ....

Louis S.J., Rawlins G.J., "Designer Genetic Algorithms: Genetic Algorithms In Structure Design", Proceedings Of The Fourth International Conference On Genetic Algorithms, San Diego, USA, July 1991, pp. 53-60.

....(1990) The previously defined processes of combination and mutation are similar to the evolutionary processes which operate at the genotype level. The computational field of genetic algorithms (Goldberg, 1989) deals with this topic. Although genetic algorithms have been used to design structures (Louis and Rawlings, 1991), their use has been for routine design. The formulations have inhibited the development of creative designs by utilising highly constrained views of what could be represented, and of combination and mutation. The notion of the genotype as a fixed length string encoding covert structure needs to ....

Louis, S. and Rawlings, G. J. (1991). Designer genetic algorithms: genetic algorithms in structure design, in R. K. Belew and L. B. Booker (eds), Proc. Fourth Int. Conf. on Genetic Algorithms, Morgan Kaufmann, San Mateo, pp. 53-60.

.... Traditionally, GAs have been used as optimizers, for example, optimizing multiparameter mathematical functions (De Jong 1975) but lately, a growing number of people have begun to use GAs more as builders than as optimizers (e.g. de Garis 1990, 1991, 1992; Koza 1990, 1992; Spofford and Hintz 1991; Louis and Rawlins 1991 etc. Thus if GAs are considered to be tools, then GP is a particular way of using such tools, namely to build evolve complex systems. The feature that makes a GA so attractive to GP is what the author calls its complexity independence (within certain limits) meaning that so long as the ....

....5.0 SOME RELATED WORK X.5. 1 Koza s Genetic Programming A growing number of researchers are now using GP techniques to build evolve increasingly complex and interesting systems, although they may not use the term GP (e.g. de Garis 1990, 1991, 1992; Koza 1990, 1992; Spofford and Hintz 1991; Louis and Rawlins 1991). One person who does, is John Koza of Stanford University, who started using the term Genetic Programming independently of and more or less simultaneously with the author. Actually Koza was in print with the term 6 months later (Koza 1990) than the author (de Garis 1990b) Koza s definition ....

[Article contains additional citation context not shown here]

Louis, S.J. and Rawlins, G.J.E., "Designer Genetic Algorithms : Genetic Algorithms in Structure Design", Proceedings of the 4th Int. Conf. on Genetic Algorithms, R.K. Belew and L.B. Booker (Eds.), Morgan Kaufmann, 1991.

....computing curricula [4] Standard graphical design aids such as Karnaugh Maps [5, 6] are widely used and tools suitable for computer implementation have evolved from the Quine McCluskey Method [7, 8] to freely available tools such as Espresso [9] and MisII [10] and many commercial products. Louis [11] is one of few sources found in the literature to address the use of GAs for the combinational logic design problem. In his dissertation [12] Louis combines knowledgebased systems with the genetic algorithm, making use of a genetic operator called masked crossover that adapts to the encoding, ....

....many XOR gates, often disregarding the NOT gate. 4 USING THE GENETIC ALGORITHM The first interesting aspect of this problem is the encoding of solutions as chromosomic strings that the GA can evolve. The representation chosen for our work is a bidimensional matrix as the one suggested by Louis [11] in which each matrix element is a gate (there are 5 types of gates: AND, NOT, OR, XOR and WIRE) that receives its 2 inputs from any 5 Input 1 Input 2 Gate Type Figure 2: Encoding used for each of the matrix elements that represent a circuit. gate y at the previous column as shown in Figure ....

[Article contains additional citation context not shown here]

Louis, S. and Rawlins, G. Designer genetic algorithms: Genetic algorithms in structure design, In Belew, R. K. and Booker, L. B. (eds) (1991) Proceedings of the Fourth International Conference on Genetic Algorithms, pp. 53--60, San Mateo, California, Morgan Kaufmann Publishers.

....CO designed to work with order dependent production programs. In contrast to classic CO, which determines corresponding crosspoints according to their respective positions in the strings, analogous CO uses the phenotypic function of parameters as the corresponding cross point criterion. Masked CO [98] Masked CO uses binary masks to direct CO. Masked CO is used to preserve schemata identified by the masks. Positionbased CO [95] A set of positions is random selected, but in this operator, the positions of elements selected in one parent are imposed on the corresponding elements in the other ....

Louis, S. J., Rawlins, J. E., "Designer Genetic Algorithms : Genetic Algorithms in Structure Design," Proc. of the 4th Int. Conf. on Genetic Algorithms, San Diego, CA, July 1991, pp. 53-60.

....area. Examples include cell placement [2] channel routing [3] test pattern generation [4] and design for test [5] and VLSI based signal processing [6] However, the use of genetic algorithms for higher level structural VLSI design synthesis has been for severely restricted forms of structure [7]. The need for this restriction arises mainly from the complexity of the solution space which grows exponentially with the number of parameters to be optimised. Such a solution space is exemplified in the case of a GA which manipulates typical VLSI cell libraries which incorporate hundreds of ....

....as individual gate delays, their constituent transistor parameters, and the length of the interconnections in the circuit and their fan in fan out. This paper reports on a GA which overcomes the restrictions of the work carried out to date in the field of structural design synthesis (for example, [7]) and that carried out at a much higher level [8] The developed GA can synthesise an unrestricted range of circuits using a multiple constrained fitness function, which considers aspects at different levels over the design hierarchy, and can deal with multiinput multi output circuits. 2. CELL ....

[Article contains additional citation context not shown here]

Louis S. J. and Rawlins G. J., "Designer Genetic Algorithms: Genetic Algorithms in Structure Design", Proceedings of the fourth international conference on Genetic Algorithms, San Diego, USA, July 1991, pp. 53-60.

....allowed to occur. They accomplished this by appending a crossover bitmap to the end of the encoded solution. The extra bitmap underwent the same crossover as the corresponding alleles thereby evolving the crossover positions. Inferior children were discarded along with their crossover bitmaps. Louis and Rawlins (1991) proposed Masked Crossover, which uses binary masks to direct crossover. Relative fitness information is translated into the binary mask, thus guiding crossover towards local fitness increases. White and Oppacher (1994) proposed Adaptive Uniform Crossover where each bit string in the population is ....

Louis, S.J. & Rawlins, G.J.E. (1991) Designer Genetic Algorithms: Genetic Algorithms in Structure Design.

....at particular locations in a GA string. Molecular biology literature makes clear that certain locations (hot spots) in DNA are much more likely than others to undergo crossover [3, 11] This issue has been addressed by several GA researchers. Punctuated crossover [18] and masked crossover [16] used masks to control whether crossover can occur at any given point. Punctuated crossover allowed the mask to be operated on like any other GA string; unfortunately the results were inconclusive. Masked crossover derived masks on the basis of the relative fitness of parents and offspring in an ....

Louis, S. J. & Rawlins, J. E. (1991). Designer Genetic Algorithms: Genetic Algorithms in Structure Design. Proceedings of the Fourth ICGA, 5360. Kaufmann, San Mateo, CA.

....was allowed to occur. They accomplished this by appending a crossover bitmap to the end of the encoded solution. The extra bitmap underwent the same crossover as the corresponding alleles thereby evolving the crossover positions. Inferior children were discarded along with their crossover bitmaps. Louis and Rawlins (1991) proposed Masked Crossover, which uses binary masks to direct crossover. Relative fitness information is translated into the binary mask, thus guiding crossover towards local fitness increases. White and Oppacher (1994) proposed Adaptive Uniform Crossover where each bit string in the population ....

Louis, S.J. & Rawlins, G.J.E. (1991) Designer Genetic Algorithms: Genetic Algorithms in Structure Design.

....the problem complexity) and so gives a more robust performance. 2.4 Other crossover techniques Many other techniques have been suggested. The idea that crossover should be more probable at some string positions than others has some basis in nature, and several such methods have been described [SM87, Hol87, Dav91a, Lev91, LR91]. The general principle is that the GA adaptively learns which sites should be favoured for crossover. This information is recorded in a punctuation string , which is itself part of the chromosome, and so is crossed over and passed on to descendants. In this way, punctuation strings which lead to ....

S.J. Louis and G.J.E. Rawlins. Designer genetic algorithms: Genetic algorithms in structure design. In R.K. Belew and L.B. Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60. Morgan Kaufmann, 1991.

....adders, the function is addition and the target technology is combinational logic gates such as boolean AND and OR gates. For parity checkers the function is parity checking and it uses the same set of logic gates. A genetic algorithm can be used to design combinational circuits as described in [8]. A five bit parity checker is one of 2 = 2 32 = 4294967296 different five input one output combinational circuits (boolean functions) If we are trying to solve this class of problems, one way of indexing (defining similarity of problems) can be as follows: Concatenate the output bit for ....

....the 2 bit adder) would have a maximum fitness of 2 4 3 = 48. A solution similarity metric depends on how solutions circuits are encoded. The phenotype is a two dimensional array of logic gates with inputs being fed into the first column and outputs being measured on the last column. [8] gives details on our representation. We use four (4) bits to encode each of 2 1 = 16 possible two input, one output gates. An additional bit specifies the location of the second input. The solution similarity metric needs to measure the distance between encoded circuits (solutions) Since we ....

S. J. Louis and G. J. E. Rawlins. Designer genetic algorithms: Genetic algorithms in structure design. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60. Morgan Kauffman, San Mateo, CA, 1991.

....design an artifact that performs the function subject to constraints. For parity checkers, the function is parity checking and the target technology is combinational logic gates such as boolean AND and OR gates. A genetic algorithm can be used to design combinational circuits as described in [12]. A fivebit parity checker is one of 2 = 2 32 = 4294967296 different five input one output combinational circuits (boolean functions) If we are trying to solve this class of problems, one way of indexing, defining similarity of problems, can be: Concatenate the output bit for all possible ....

Sushil J. Louis and Gregory J. E. Rawlins. Designer genetic algorithms: Genetic algorithms in structure design. In Proceedings of the Fourth pages 53--60. Morgan Kauffman, San Mateo, CA, 1991.

....for the two robot tracks. For each gate, four bits are needed for expressing the 16 possible logic gates with two inputs and one output. Therefore the chromosome length will be 36 Theta 4 = 144. We map a two dimensional logic circuit to a one dimensional chromosome by concatenating adjacent rows [8]. The simulation process provides an evaluation of a candidate combinational circuit for controlling the robot. The encoded chromosome of the combinational logical gate is obtained from the GA and evaluated in the simulation environment using a fitness function that measures how well the simbot ....

Sushil J. Louis and Gregory J. E. Rawlins. Designer genetic algorithms: Genetic algorithms in structure design. In R. Belew and L. Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60. Morgan Kauffman, San Mateo, CA, 1991.

....and flexibility and they are especially suitable for searching large, complex, and poorly understand spaces often encountered in design tasks. During the last decade, many papers have been written to explore the evolution of robot control strategies. Evolving combinational circuits was reported by Louis and Rawlins in 1991 (Louis 91) Louis, McGraw, and Wyckoff applied Case Based Reasoning (CBR) to GAs as an analysis tool for the parity combinational circuit design problem (Loius 92) A series of technical reports has been published by Cliff, Husbands, and Harvey about using genetic algorithms to design ....

S. J. Louis and G. J. E. Rawlins. "Designer genetic algorithms: Genetic algorithms in structure design." In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60. Morgan Kauffman, San Mateo, CA, 1991.

....a problem and breaking it into simpler parts; the genetic algorithm does what it is good at combining subsolutions into more complete solutions. Recent work in this field has explored the evolution of structure and control [3, 4, 7] while evolving combinational circuits is described in [8]. Braitenberg s book elegantly outlines the emergence of complex behavior from simpler ones [1] The next section describes genetic algorithms. Section 3 explains the agent, environment, and learning task while section 4 discusses the genetic algorithm encoding. Section 5 provides some results and ....

Sushil J. Louis and Gregory J. E. Rawlins. Designer genetic algorithms: Genetic algorithms in structure design. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60. Morgan Kauffman, San Mateo, CA, 1991.

No context found.

Louis, S. J. and Rawlins, G. J. E.. "Designer genetic algorithms: Genetic algorithms in structure design." In Proceedings of the Fourth International Conference on Genetic Algorithms. pp. 53-60. Morgan Kauffmann Publishers, Inc., 1991.

....AND AVERAGE HAMMING DISTANCE Assuming that offspring replace their parents during a crossover, all crossover operators can be partitioned into two groups based on whether or not they change the hamming average. If one parent contributes the same alleles to both offspring (as in masked crossover (Louis and Rawlins 1991)) the hamming distance between the children is less than the hamming distance between their parents. This leads to a loss of genetic material, reducing population hamming average and resulting in faster hamming convergence. We do not consider such operators in this paper. The vast majority of ....

....consisting of mixed deceptive and easy subproblems the GAC still does much better than the classical GA. Finally, GA hard problems, that are both deceptive and epistatic, may need masked crossover or other length independent recombinant operators to be solved successfully using complements (Louis and Rawlins 1991). 5 PREDICTING TIME TO CONVERGENCE We now have an upper bound on time to convergence on static functions, but predicting performance on an arbitrary function is more difficult because of the non linearities introduced by the selection intensity. However, computing the rate of decrease in the ....

Louis, Sushil J., and Rawlins, Gregory J. E. "Designer Genetic Algorithms: Genetic Algorithms in Structures Design," Proceedings of the Fourth International Conference on Genetic Algorithms, Morgan Kauffman, 1991, 53-60.

....sufficient precision for plasma diagnostic purposes. Figure 4 illustrates the encoding and mapping of our parameters. We expect our proposed method to work for more complete physical models and therefore implemented an elitist selection scheme modeled after the CHC algorithm with linear scaling (Louis and Rawlins, 1991) that is claimed to converge quickly. A combination of conservative selection, highly destructive crossover 400 600 800 1000 1200 5.0x10 23 1.0x10 24 1.5x10 24 2.0x10 24 1x10 3 1x10 4 Fitness Temperature (eV) Density (cm 3 ) Figure 3: Fitness as a function of two fitting ....

Louis, S. J. and Rawlins, G. J. E. (1991). Designer genetic algorithms: Genetic algorithms in structure design. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60.

No context found.

S. J. Louis and G. J. E. Rawlins. "Designer genetic algorithms: Genetic algorithms in structure design." In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 53--60. Morgan Kauffman, San Mateo, CA, 1991.

No context found.

Louis, S. J. and Rawlins, G. J. E. (1991) Designer Genetic Algorithms: Genetic Algorithms in Structures Design. In R. K. Belew and L. B. Booker (eds.) Proceedings of the Fourth International Conference on Genetic Algorithms (San Mateo: Morgan Kauffman), 53-60.

No context found.

S. J. Louis and G. J. Rawlins. Designer Genetic Algorithms: Genetic Algorithms in Structure Design. In Proceedings of the 4th International Conference on Genetic Algorithms, pages 53--60, San Diego California, . 1991. Morgan Kaufmann Publishiers, Inc.

No context found.

Louis, S. & Rawlings, G.J. Designer genetic algorithms: genetic algorithms in structure design, in Proc. Fourth Int. Conf. on Genetic Algorithms, eds. R.K. Belew & L.B. Booker, Morgan Kaufmann, San Mateo, 1991, pp.53-60.

No context found.

Louis, S. and Rawlings, G. J. (1991). Designer genetic algorithms: genetic algorithms in structure design, in R. K. Belew and L. B. Booker (eds), Proc. Fourth Int. Conf. on Genetic Algorithms, Morgan Kaufmann, San Mateo, pp.53-60.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC