113:
of existing structures is done at a smaller timescale (structural exploitation). In the structural exploration phase, new neural structures are developed by gradually adding new structures to an initially minimal network that is used as a starting point. In the structural exploitation phase, the
176:
Yohannes
Kassahun, Mark Edgington, Jan Hendrik Metzen, Gerald Sommer and Frank Kirchner. Common Genetic Encoding for Both Direct and Indirect Encodings of Networks. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2007), London, UK, 1029–1036,
235:
Jan
Hendrik Metzen, Mark Edgington, Yohannes Kassahun and Frank Kirchner. Performance Evaluation of EANT in the RoboCup Keepaway Benchmark. In Proceedings of the Sixth International Conference on Machine Learning and Applications (ICMLA 2007), pages 342–347, USA, 2007
258:
Nils T Siebel and Gerald Sommer. Learning Defect
Classifiers for Visual Inspection Images by Neuro-evolution using Weakly Labelled Training Data. Proceedings of the IEEE Congress on Evolutionary Computation (CEC 2008), pages 3926–3932, Hong Kong, China, June 2008.
188:
Yohannes
Kassahun and Gerald Sommer. Efficient reinforcement learning through evolutionary acquisition of neural topologies. In Proceedings of the 13th European Symposium on Artificial Neural Networks (ESANN 2005), pages 259–266, Bruges, Belgium, April 2005.
79:(CGE) that handles both direct and indirect encoding of neural networks within the same theoretical framework. The encoding has important properties that makes it suitable for evolving neural networks:
130:
keepaway benchmark. In all the tests, EANT was found to perform very well. Moreover, a newer version of EANT, called EANT2, was tested on a visual servoing task and found to outperform
247:
Nils T Siebel and Gerald Sommer. Evolutionary reinforcement learning of artificial neural networks. International
Journal of Hybrid Intelligent Systems 4(3): 171–183, October 2007.
154:
Peter J Angeline, Gregory M Saunders, and Jordan B Pollack. An evolutionary algorithm that constructs recurrent neural networks. IEEE Transactions on Neural
Networks, 5:54–65, 1994.
44:. It is closely related to the works of Angeline et al. and Stanley and Miikkulainen. Like the work of Angeline et al., the method uses a type of parametric mutation that comes from
222:
260:
131:
57:
290:
191:
56:
in EANT2), in which adaptive step sizes are used for optimizing the weights of the neural networks. Similar to the work of
Stanley (
300:
68:
Despite sharing these two properties, the method has the following important features which distinguish it from previous works in
295:
248:
135:
237:
41:
49:
34:
37:
275:
105:
For evolving the structure and weights of neural networks, an evolutionary process is used, where the
126:
EANT has been tested on some benchmark problems such as the double-pole balancing problem, and the
166:
115:
45:
19:
This article is on evolutionary acquisition of artificial neural topologies, not of natural ones.
216:
198:
60:), the method starts with minimal structures which gain complexity along the evolution path.
69:
94:, i.e. every valid genotype represents a valid phenotype. (Similarly, the encoding is
284:
109:
of structures is executed at a larger timescale (structural exploration), and the
178:
165:
NeuroEvolution of
Augmented Topologies (NEAT) by Stanley and Miikkulainen, 2005
155:
16:
Method that evolves both the topology and weights of artificial neural networks
138:
method. Further experiments include results on a classification problem
127:
87:
in that it is able to represent all types of valid phenotype networks.
114:
weights of the currently available structures are optimized using an
53:
52:(now using the most advanced form of the evolution strategies
40:
method that evolves both the topology and weights of
23:Evolutionary acquisition of neural topologies
8:
102:These properties have been formally proven.
98:such as structural mutation and crossover.)
75:It introduces a genetic encoding called
147:
221:: CS1 maint: archived copy as title (
214:
64:Contribution of EANT to neuroevolution
7:
276:BEACON Blog: What is neuroevolution?
14:
134:and the traditional iterative
96:closed under genetic operators
1:
317:
291:Artificial neural networks
42:artificial neural networks
18:
301:Evolutionary computation
50:evolutionary programming
296:Evolutionary algorithms
77:common genetic encoding
38:reinforcement learning
46:evolution strategies
116:evolution strategy
308:
263:
256:
250:
245:
239:
233:
227:
226:
220:
212:
210:
209:
203:
197:. Archived from
196:
186:
180:
174:
168:
163:
157:
152:
316:
315:
311:
310:
309:
307:
306:
305:
281:
280:
272:
267:
266:
257:
253:
246:
242:
234:
230:
213:
207:
205:
201:
194:
192:"Archived copy"
190:
187:
183:
175:
171:
164:
160:
153:
149:
144:
124:
66:
20:
17:
12:
11:
5:
314:
312:
304:
303:
298:
293:
283:
282:
279:
278:
271:
270:External links
268:
265:
264:
251:
240:
228:
181:
169:
158:
146:
145:
143:
140:
123:
120:
100:
99:
88:
70:neuroevolution
65:
62:
15:
13:
10:
9:
6:
4:
3:
2:
313:
302:
299:
297:
294:
292:
289:
288:
286:
277:
274:
273:
269:
261:
255:
252:
249:
244:
241:
238:
232:
229:
224:
218:
204:on 2007-06-13
200:
193:
185:
182:
179:
173:
170:
167:
162:
159:
156:
151:
148:
141:
139:
137:
133:
129:
121:
119:
117:
112:
108:
103:
97:
93:
89:
86:
82:
81:
80:
78:
73:
71:
63:
61:
59:
55:
51:
47:
43:
39:
36:
32:
28:
24:
254:
243:
231:
206:. Retrieved
199:the original
184:
172:
161:
150:
136:Gauss–Newton
125:
111:exploitation
110:
106:
104:
101:
95:
91:
84:
76:
74:
67:
35:evolutionary
30:
26:
22:
21:
122:Performance
107:exploration
285:Categories
208:2008-02-11
142:References
217:cite web
85:complete
33:) is an
128:RoboCup
92:closed
90:It is
83:It is
54:CMA-ES
202:(PDF)
195:(PDF)
177:2007.
31:EANT2
223:link
132:NEAT
58:NEAT
48:and
27:EANT
287::
219:}}
215:{{
118:.
72:.
262:.
225:)
211:.
29:/
25:(
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.