下载文档原格式
武汉工业学院毕业论文论文题目：智能信息处理理论与方法姓名王斌学号 *********院（系）数理科学系专业电子信息科学与技术指导教师李相虎2011年6月10日目录第一章绪论 (1)1.1智能信息处理的产生及其发展 (1)1.1.1 计算智能的产生 (1)1.1.2 信息处理技术的应用和现状 (4)第二章智能信息处理的主要技术 (8)2.1.1 什么是模糊逻辑 (8)2.1.2 模糊逻辑控制技术 (10)2.1.3 模糊处理技术 (12)2.2 神经计算技术 (13)2.2.1 脑神经系统 (13)2.2.2 神经网络的主要特征 (14)2.3 进化计算技术 (15)2.3.1遗传算法的发展过程 (15)2.3.2 遗传算法的基本理论研究 (15)2.3.3 进化计算与遗传算法的关系 (16)2.3.4 遗传算法参数的选择 (16)2.3.5 遗传算法的应用 (17)2.4 混沌计算技术 (19)2.4.1 混沌时间序列预测和控制 (20)2.4.2 混沌神经网络 (20)2.4.3 混沌在多Agent系统中的应用 (21)2.4.4混沌同步和通信 (22)2.5 分形计算技术 (24)2.5.1分形的基本概念 (24)2.5.2 分形维数 (26)2.5.3 分形的应用 (26)第三章总结与展望 (27)致谢 (28)参考文献 (29)摘要随着信息技术在企业的日益普及，信息系统在工具手段、开发模式、软件规模、指导思想等方面的不断提升，企业在信息需求方面日趋多样化、人性化、智能化，智能信息处理成为信息系统发展的一个重要方向。
多年来，人们一直在探索新一代的信息处理技术。
自20世纪90年代以来，国际上掀起了一股强劲的研究模糊逻辑系统、神经网络、遗传算法、信息融合、混沌与分形理论与技术的热潮，推动了软计算、软处理技术的深入发展。
近年来，模糊计算、神经计算、进化计算、混沌与分形计算、小波交换、人工生命科学等新一代智能信息处理技术的研究，不仅在各自的学科领域取得了引人瞩目的发展，而且它们之间的相互渗透和有机结合必然引起智能信息处理技术的革命。
刊名简称刊名全称ISSN小类名称（中文）小类名称（英文）小类分区HUM-COMPUT I HUMAN-COMPUTER I0737-0024计算机：控制论COMPUTER SCIENCE, CYBER1 IEEE T SYST IEEE TRANSACTION1083-4419计算机：控制论COMPUTER SCIENCE, CYBER2INT J HUM-CO INTERNATIONAL JO1071-5819计算机：控制论COMPUTER SCIENCE, CYBER2 USER MODEL U USER MODELING AN0924-1868计算机：控制论COMPUTER SCIENCE, CYBER2 ACM T COMPUT ACM Transactions1073-0516计算机：控制论COMPUTER SCIENCE, CYBER3 BIOL CYBERN B IOLOGICAL CYBER0340-1200计算机：控制论COMPUTER SCIENCE, CYBER3 IEEE T SYST IEEE TRANSACTION1083-4427计算机：控制论COMPUTER SCIENCE, CYBER3 IEEE T SYST IEEE TRANSACTION1094-6977计算机：控制论COMPUTER SCIENCE, CYBER3 MACH VISION MACHINE VISION A0932-8092计算机：控制论COMPUTER SCIENCE, CYBER3 BEHAV INFORM BEHAVIOUR & INFO0144-929X计算机：控制论COMPUTER SCIENCE, CYBER4 CONTROL CYBE CONTROL AND CYBE0324-8569计算机：控制论COMPUTER SCIENCE, CYBER4 CYBERNET SYS CYBERNETICS AND 0196-9722计算机：控制论COMPUTER SCIENCE, CYBER4INT J HUM-CO INTERNATIONAL JO1044-7318计算机：控制论COMPUTER SCIENCE, CYBER4 INTERACT COM INTERACTING WITH0953-5438计算机：控制论COMPUTER SCIENCE, CYBER4J COMPUT SYS JOURNAL OF COMPU1064-2307计算机：控制论COMPUTER SCIENCE, CYBER4 KYBERNETESKYBERNETES0368-492X计算机：控制论COMPUTER SCIENCE, CYBER4 KYBERNETIKA K YBERNETIKA0023-5954计算机：控制论COMPUTER SCIENCE, CYBER4 MODEL IDENT MODELING IDENTIF0332-7353计算机：控制论COMPUTER SCIENCE, CYBER4 PRESENCE-TEL PRESENCE-TELEOPE1054-7460计算机：控制论COMPUTER SCIENCE, CYBER4 IEEE T EVOLU IEEE TRANSACTION1089-778X计算机：人工智能COMPUTER SCIENCE, ARTIF1 IEEE T FUZZY IEEE TRANSACTION1063-6706计算机：人工智能COMPUTER SCIENCE, ARTIF1 IEEE T PATTE IEEE TRANSACTION0162-8828计算机：人工智能COMPUTER SCIENCE, ARTIF1INT J COMPUT INTERNATIONAL JO0920-5691计算机：人工智能COMPUTER SCIENCE, ARTIF1 SIAM J IMAGI SIAM Journal on 1936-4954计算机：人工智能COMPUTER SCIENCE, ARTIF1 ADV ENG INFO ADVANCED ENGINEE1474-0346计算机：人工智能COMPUTER SCIENCE, ARTIF2 APPL SOFT CO APPLIED SOFT COM1568-4946计算机：人工智能COMPUTER SCIENCE, ARTIF2 ARTIF INTELL ARTIFICIAL INTEL0004-3702计算机：人工智能COMPUTER SCIENCE, ARTIF2 ARTIF LIFEARTIFICIAL LIFE1064-5462计算机：人工智能COMPUTER SCIENCE, ARTIF2 COMPUT INTEL COMPUTATIONAL IN0824-7935计算机：人工智能COMPUTER SCIENCE, ARTIF2 COMPUT LINGU COMPUTATIONAL LI0891-2017计算机：人工智能COMPUTER SCIENCE, ARTIF2 COMPUT VIS I COMPUTER VISION 1077-3142计算机：人工智能COMPUTER SCIENCE, ARTIF2 DECIS SUPPOR DECISION SUPPORT0167-9236计算机：人工智能COMPUTER SCIENCE, ARTIF2 EVOL COMPUT E VOLUTIONARY COM1063-6560计算机：人工智能COMPUTER SCIENCE, ARTIF2 EXPERT SYST EXPERT SYSTEMS W0957-4174计算机：人工智能COMPUTER SCIENCE, ARTIF2 IEEE COMPUT IEEE Computation1556-603X计算机：人工智能COMPUTER SCIENCE, ARTIF2 IEEE INTELL IEEE INTELLIGENT1541-1672计算机：人工智能COMPUTER SCIENCE, ARTIF2 IEEE T IMAGE IEEE TRANSACTION1057-7149计算机：人工智能COMPUTER SCIENCE, ARTIF2 IEEE T KNOWL IEEE TRANSACTION1041-4347计算机：人工智能COMPUTER SCIENCE, ARTIF2 IEEE T NEURA IEEE TRANSACTION1045-9227计算机：人工智能COMPUTER SCIENCE, ARTIF2 IEEE T SYST IEEE TRANSACTION1083-4419计算机：人工智能COMPUTER SCIENCE, ARTIF2 INFORM FUSIO Information Fusi1566-2535计算机：人工智能COMPUTER SCIENCE, ARTIF2INT J INNOV International Jo1349-4198计算机：人工智能COMPUTER SCIENCE, ARTIF2INT J NEURAL International Jo0129-0657计算机：人工智能COMPUTER SCIENCE, ARTIF2 INTEGR COMPU INTEGRATED COMPU1069-2509计算机：人工智能COMPUTER SCIENCE, ARTIF2J MACH LEARN JOURNAL OF MACHI1532-4435计算机：人工智能COMPUTER SCIENCE, ARTIF2J WEB SEMANT Journal of Web S1570-8268计算机：人工智能COMPUTER SCIENCE, ARTIF2 MED IMAGE AN MEDICAL IMAGE AN1361-8415计算机：人工智能COMPUTER SCIENCE, ARTIF2 NEURAL COMPU NEURAL COMPUTATI0899-7667计算机：人工智能COMPUTER SCIENCE, ARTIF2 PATTERN RECO PATTERN RECOGNIT0031-3203计算机：人工智能COMPUTER SCIENCE, ARTIF2 ACM T AUTON ACM Transactions1556-4665计算机：人工智能COMPUTER SCIENCE, ARTIF3AUTON AGENT AUTONOMOUS AGENT1387-2532计算机：人工智能COMPUTER SCIENCE, ARTIF3 AUTON ROBOT A UTONOMOUS ROBOT0929-5593计算机：人工智能COMPUTER SCIENCE, ARTIF3 CHEMOMETR IN CHEMOMETRICS AND0169-7439计算机：人工智能COMPUTER SCIENCE, ARTIF3 CONSTRAINTS C ONSTRAINTS1383-7133计算机：人工智能COMPUTER SCIENCE, ARTIF3 DATA KNOWL E DATA & KNOWLEDGE0169-023X计算机：人工智能COMPUTER SCIENCE, ARTIF3 DATA MIN KNO DATA MINING AND 1384-5810计算机：人工智能COMPUTER SCIENCE, ARTIF3 ENG APPL ART ENGINEERING APPL0952-1976计算机：人工智能COMPUTER SCIENCE, ARTIF3 IEEE T SYST IEEE TRANSACTION1094-6977计算机：人工智能COMPUTER SCIENCE, ARTIF3 IMAGE VISION IMAGE AND VISION0262-8856计算机：人工智能COMPUTER SCIENCE, ARTIF3 INT J APPROX INTERNATIONAL JO0888-613X计算机：人工智能COMPUTER SCIENCE, ARTIF3 INT J INF TE INTERNATIONAL JO0219-6220计算机：人工智能COMPUTER SCIENCE, ARTIF3 INT J SEMANT International Jo1552-6283计算机：人工智能COMPUTER SCIENCE, ARTIF3 J AMB INTEL Journal of Ambie1876-1364计算机：人工智能COMPUTER SCIENCE, ARTIF3 J ARTIF INTE JOURNAL OF ARTIF1076-9757计算机：人工智能COMPUTER SCIENCE, ARTIF3 J AUTOM REAS JOURNAL OF AUTOM0168-7433计算机：人工智能COMPUTER SCIENCE, ARTIF3 J HEURISTICS JOURNAL OF HEURI1381-1231计算机：人工智能COMPUTER SCIENCE, ARTIF3 J INTELL MAN JOURNAL OF INTEL0956-5515计算机：人工智能COMPUTER SCIENCE, ARTIF3 KNOWL INF SY KNOWLEDGE AND IN0219-1377计算机：人工智能COMPUTER SCIENCE, ARTIF3 MACH LEARNMACHINE LEARNING0885-6125计算机：人工智能COMPUTER SCIENCE, ARTIF3 MACH VISION MACHINE VISION A0932-8092计算机：人工智能COMPUTER SCIENCE, ARTIF3 NETWORK-COMP NETWORK-COMPUTAT0954-898X计算机：人工智能COMPUTER SCIENCE, ARTIF3 NEURAL NETWO NEURAL NETWORKS0893-6080计算机：人工智能COMPUTER SCIENCE, ARTIF3 SOFT COMPUT S OFT COMPUTING1432-7643计算机：人工智能COMPUTER SCIENCE, ARTIF3 ADV ELECTR C Advances in Elec1582-7445计算机：人工智能COMPUTER SCIENCE, ARTIF4 AI COMMUN AI COMMUNICATION0921-7126计算机：人工智能COMPUTER SCIENCE, ARTIF4 AI EDAM AI EDAM-ARTIFICI0890-0604计算机：人工智能COMPUTER SCIENCE, ARTIF4 AI MAG AI MAGAZINE0738-4602计算机：人工智能COMPUTER SCIENCE, ARTIF4 ANN MATH ART ANNALS OF MATHEM1012-2443计算机：人工智能COMPUTER SCIENCE, ARTIF4 APPL ARTIF I APPLIED ARTIFICI0883-9514计算机：人工智能COMPUTER SCIENCE, ARTIF4 APPL INTELL A PPLIED INTELLIG0924-669X计算机：人工智能COMPUTER SCIENCE, ARTIF4 ARTIF INTELL ARTIFICIAL INTEL0933-3657计算机：人工智能COMPUTER SCIENCE, ARTIF4 ARTIF INTELL ARTIFICIAL INTEL0269-2821计算机：人工智能COMPUTER SCIENCE, ARTIF4 COGN SYST RE Cognitive System1389-0417计算机：人工智能COMPUTER SCIENCE, ARTIF4 COMPUT INFOR COMPUTING AND IN1335-9150计算机：人工智能COMPUTER SCIENCE, ARTIF4 COMPUT SPEEC COMPUTER SPEECH 0885-2308计算机：人工智能COMPUTER SCIENCE, ARTIF4 CONNECT SCI C ONNECTION SCIEN0954-0091计算机：人工智能COMPUTER SCIENCE, ARTIF4 EXPERT SYST E XPERT SYSTEMS0266-4720计算机：人工智能COMPUTER SCIENCE, ARTIF4 FUZZY OPTIM Fuzzy Optimizati1568-4539计算机：人工智能COMPUTER SCIENCE, ARTIF4 GENET PROGRA Genetic Programm1389-2576计算机：人工智能COMPUTER SCIENCE, ARTIF4 IET COMPUT V IET Computer Vis1751-9632计算机：人工智能COMPUTER SCIENCE, ARTIF4 INF TECHNOL Information Tech1392-124X计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT ARAB J I International Ar1683-3198计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J AP MAT International Jo1641-876X计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J ARTIF International Jo0218-2130计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J COMPUT International Jo1875-6883计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J DOC AN International Jo1433-2833计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J FUZZY International Jo1562-2479计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J INTELL INTERNATIONAL JO0884-8173计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J PATTER INTERNATIONAL JO0218-0014计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J SOFTW INTERNATIONAL JO0218-1940计算机：人工智能COMPUTER SCIENCE, ARTIF4 INT J UNCERT INTERNATIONAL JO0218-4885计算机：人工智能COMPUTER SCIENCE, ARTIF4INTELL DATA Intelligent Data1088-467X计算机：人工智能COMPUTER SCIENCE, ARTIF4 J CHEMOMETR J OURNAL OF CHEMO0886-9383计算机：人工智能COMPUTER SCIENCE, ARTIF4 J COMPUT SYS JOURNAL OF COMPU1064-2307计算机：人工智能COMPUTER SCIENCE, ARTIF4 J EXP THEOR JOURNAL OF EXPER0952-813X计算机：人工智能COMPUTER SCIENCE, ARTIF4 J INTELL FUZ JOURNAL OF INTEL1064-1246计算机：人工智能COMPUTER SCIENCE, ARTIF4 J INTELL INF JOURNAL OF INTEL0925-9902计算机：人工智能COMPUTER SCIENCE, ARTIF4 J INTELL ROB JOURNAL OF INTEL0921-0296计算机：人工智能COMPUTER SCIENCE, ARTIF4 J MATH IMAGI JOURNAL OF MATHE0924-9907计算机：人工智能COMPUTER SCIENCE, ARTIF4 J MULT-VALUE JOURNAL OF MULTI1542-3980计算机：人工智能COMPUTER SCIENCE, ARTIF4 J REAL-TIME Journal of Real-1861-8200计算机：人工智能COMPUTER SCIENCE, ARTIF4 KNOWL ENG RE KNOWLEDGE ENGINE0269-8889计算机：人工智能COMPUTER SCIENCE, ARTIF4 KNOWL-BASED KNOWLEDGE-BASED 0950-7051计算机：人工智能COMPUTER SCIENCE, ARTIF4 MALAYS J COM Malaysian Journa0127-9084计算机：人工智能COMPUTER SCIENCE, ARTIF4 MECHATRONICS MECHATRONICS0957-4158计算机：人工智能COMPUTER SCIENCE, ARTIF4 MIND MACH MINDS AND MACHIN0924-6495计算机：人工智能COMPUTER SCIENCE, ARTIF4 NEURAL COMPU NEURAL COMPUTING0941-0643计算机：人工智能COMPUTER SCIENCE, ARTIF4 NEURAL NETW Neural Network W1210-0552计算机：人工智能COMPUTER SCIENCE, ARTIF4 NEURAL PROCE NEURAL PROCESSIN1370-4621计算机：人工智能COMPUTER SCIENCE, ARTIF4 NEUROCOMPUTI NEUROCOMPUTING0925-2312计算机：人工智能COMPUTER SCIENCE, ARTIF4 PATTERN ANAL PATTERN ANALYSIS1433-7541计算机：人工智能COMPUTER SCIENCE, ARTIF4 PATTERN RECO PATTERN RECOGNIT0167-8655计算机：人工智能COMPUTER SCIENCE, ARTIF4 ROBOT AUTON ROBOTICS AND AUT0921-8890计算机：人工智能COMPUTER SCIENCE, ARTIF4 TRAIT SIGNAL Traitement du Si0765-0019计算机：人工智能COMPUTER SCIENCE, ARTIF4 TURK J ELECT Turkish Journal 1300-0632计算机：人工智能COMPUTER SCIENCE, ARTIF42010年总被引频次09年总被引频大类名称大类分区2010年影响因子009年影响因子2010年平均影响008年影响因子工程技术1 4.000 6.190 2.905 4.36511751601工程技术2 2.699 3.007 2.361 2.68945224901工程技术2 1.600 2.380 1.769 1.91622452539工程技术2 3.074 2.345 1.483 2.301552672工程技术3 1.857 1.1940.000 1.526662875工程技术3 1.667 1.697 1.935 1.76644894624工程技术3 2.093 2.033 1.350 1.82522922379工程技术3 2.105 2.016 1.375 1.83213901507工程技术3 1.5490.952 1.485 1.329702574工程技术40.8350.7670.9150.839824873工程技术40.3000.3780.6890.456454415工程技术40.6620.7800.4940.645363534工程技术40.6810.5870.3210.530425534工程技术3 1.192 1.698 1.103 1.331821980工程技术40.1910.1680.0820.14712793工程技术40.3030.3080.2350.282441544工程技术40.4610.4450.2810.396463526工程技术40.4070.4350.0000.421158110工程技术4 1.097 1.2410.750 1.029********工程技术1 4.403 4.589 3.736 4.24346815038工程技术2 2.695 3.343 3.624 3.22148755961工程技术1 5.308 4.378 5.960 5.2152506022851工程技术1 5.151 3.508 5.358 4.67298987324工程技术1 4.5000.0000.000 4.5003220工程技术3 1.400 1.730 1.848 1.659495460工程技术2 2.097 2.415 1.909 2.14012061079工程技术2 2.533 3.036 3.397 2.98962659016工程技术3 2.143 1.960 1.164 1.7568161070工程技术20.704 5.378 3.310 3.1314081223工程技术2 2.971 2.212 2.656 2.61313201467工程技术2 2.534 1.676 2.220 2.14328592369工程技术2 2.135 2.622 1.873 2.21032333376工程技术2 2.652 3.103 3.000 2.91819922397工程技术2 1.926 2.908 2.596 2.47766155421工程技术2 2.905 2.622 2.535 2.687385283工程技术2 2.570 3.144 2.278 2.66416612214工程技术2 2.918 2.848 3.315 3.0271277411472工程技术2 1.851 2.285 2.236 2.12436544262工程技术2 2.633 2.889 3.726 3.0839********工程技术2 2.699 3.007 2.361 2.68945224901工程技术2 1.621 2.212 2.057 1.963620536工程技术2 1.667 2.932 2.791 2.46319361690工程技术2 4.237 2.9880.901 2.709816669工程技术3 2.122 2.0420.617 1.594304254工程技术2 2.974 2.789 3.116 2.96047664132工程技术2 2.789 3.412 3.023 3.075578651工程技术1 4.364 3.093 3.602 3.68629012307工程技术2 2.290 2.175 2.378 2.28175817456工程技术2 2.682 2.554 3.279 2.8381172611703工程技术3 1.000 1.3640.000 1.182108114工程技术2 2.103 1.510 2.125 1.913524551工程技术3 2.033 1.235 1.500 1.58913191215工程技术2 2.222 2.111 1.940 2.0913*******工程技术3 1.410 1.2970.879 1.195268307工程技术3 1.722 1.745 1.480 1.64913121398工程技术2 1.238 2.950 2.421 2.20315091822工程技术3 1.352 1.444 1.397 1.39814261483工程技术3 2.105 2.016 1.375 1.83213901507工程技术3 1.578 1.474 1.496 1.51632362926工程技术3 1.684 2.090 1.708 1.82713841686工程技术3 3.139 1.3120.953 1.801534322工程技术2 1.679 2.345 1.893 1.972123167工程技术3 1.5000.0000.000 1.500430工程技术3 1.691 1.981 1.611 1.76117392238工程技术2 2.280 1.926 1.691 1.966579522工程技术3 1.623 1.264 1.064 1.317629764工程技术4 1.0810.938 1.018 1.012854979工程技术2 2.008 2.211 1.733 1.984727773工程技术2 1.967 1.663 2.326 1.98563107305工程技术3 1.5490.952 1.485 1.329702574工程技术30.957 1.536 1.333 1.275742792工程技术2 1.972 1.879 2.656 2.16958256488工程技术3 1.512 1.3280.984 1.27510451096工程技术40.7000.5090.0000.6058347工程技术40.8370.7550.6080.733294461工程技术40.6400.6360.4770.584299357工程技术40.525 1.0180.6910.7457121180工程技术40.4300.8930.7220.682472734工程技术40.5630.5800.7950.646429621工程技术40.8930.9880.7750.885508534工程技术3 1.568 1.645 1.960 1.72411101475工程技术40.4290.0570.1190.202462592工程技术4 1.0000.571 1.581 1.051223251工程技术40.3560.4560.4920.435104131工程技术3 1.353 1.034 1.413 1.267626636工程技术4 1.0570.8060.8840.916358420工程技术40.733 1.2310.7170.894241332工程技术40.702 1.0590.0000.881147136工程技术3 1.167 1.0910.000 1.12910258工程技术40.4470.9690.6670.6943733工程技术40.6380.4950.0000.5678254工程技术40.0650.0000.0000.065250工程技术40.7940.6840.0000.739418407工程技术40.3300.4360.6670.478235244工程技术4 1.4710.3120.0000.89212410工程技术4 1.091 1.2130.909 1.071253311工程技术3 1.362 1.0900.000 1.226207215工程技术3 1.331 1.1940.860 1.12810871309工程技术40.6850.5120.6600.619698831工程技术40.2480.3270.4470.341185258工程技术40.850 1.147 1.0000.999718990工程技术40.4120.9290.4260.589427526化学4 1.377 1.291 1.415 1.36121862095工程技术40.1910.1680.0820.14712793工程技术40.6550.5330.3410.510207312工程技术40.6480.7400.6490.679318336工程技术40.8750.980 1.0750.977399545工程技术40.7570.8580.5600.725575568工程技术3 1.285 1.437 1.331 1.35110631074工程技术40.3330.3430.3080.3288570工程技术40.9620.0000.0000.9621060工程技术3 1.257 1.143 1.588 1.329635954工程技术3 1.574 1.3080.924 1.269914933工程技术40.3640.0000.0000.364210工程技术30.944 1.198 1.434 1.19211731214工程技术40.6180.7830.3400.580216266工程技术40.5630.8120.7670.714401474工程技术40.5110.4750.3950.460200180工程技术4 1.088 1.0150.942 1.015567634工程技术3 1.442 1.440 1.234 1.37240354108工程技术3 1.097 1.293 1.367 1.252433471工程技术3 1.235 1.303 1.559 1.36646984545工程技术3 1.313 1.361 1.214 1.29616651526工程技术40.0780.1210.0000.1004978工程技术40.2860.0000.0000.28681008年总被引频备注1030399118614814259189911696246064013843185846443237591838495211246748660354606712075581512862548263318163032123179612214381398833991472907493154332843822536730111493625 178 **** **** 1192 1169 3067 1249 234 78 0 1766 470 586 782 550 6211 624 666 5706 723 0 310 274 814 583 484 407 1025 482 194 95 461 373 261 0 0 9 0 0 0 204 0 204 0 1039 760 235 75064 230 338 529 497 804 49 0 701 680 0 1051 166 362 174 564 2946 402 4256 1364 0 0。
哈尔滨工业大学毕业论文打印装订规范一．论文结构及要求毕业设计（论文)由以下几部分组成：封面、内封（扉页、题目)、毕业设计（论文）评语、毕业设计(论文）任务书、中文摘要、外文摘要、目录、正文、致谢、参考文献、附录。
二．打印及装订要求1. 字体论文正文所用字体要求为宋体。
2。
字号第一层次（章）题序和标题用小二号黑体字。
第二层次（节）题序和标题用小三号黑体字。
第三层次（条）题序和标题用四号黑体字.第四层次（款)题序和标题用小四号黑体字.第五层次(项）题序和标题用小四号宋体字.正文用小四号宋体字。
页码用小五号字，在底线下居中.论文的中文和外文摘要属二次文献置于目录前，并编入目录，按第一层次(章）的编辑要求处理.致谢、参考文献、附录同样按第一层次（章）的编辑要求处理，另起新页，与正文一起顺序用阿拉伯数字编页。
3。
页眉与页脚毕业设计（论文）除封皮及内封、评语页、任务书外，各页均应加页眉，在版芯上边线隔行加粗细双线（粗线在上，线宽0。
8mm），双线上居中打印页眉文字，奇偶页页眉文字均为“哈尔滨工业大学远程教育本（专)科毕业设计(论文）”.页脚设定为各部分的页码.页码在版芯下边线之下隔行居中放置。
摘要、目录等文前部分的页码用罗马数字单独编排，正文以后的页码用阿拉伯数字编排。
4。
摘要及关键词摘要题头用小二号黑体字居中排写，然后隔行书写摘要的文字部分。
关键词题头用小四号黑体字顶格书写，然后空一格书写有关关键词，各关键词之间加标点符号；最后一词之后不加标点符号.5。
目录目录中各章题序及标题用小四号黑体，其余用小四号宋体.6。
装订顺序毕业设计(论文)按以下顺序装订:封面、内封（扉页）、毕业设计（论文）评语、毕业设计（论文）任务书、中文摘要、外文摘要、目录、正文、致谢、参考文献、附录。
装订裁剪后论文尺寸为186mm×258mm。
三。
打印规范★打印时,论文版芯大小一般应为145mm×210mm（包括页眉及页码则为145mm×230mm）,每页33行.毕业设计（论文）除封皮及内封、评语页、任务书外，各页均应加页眉，在版心上边线隔行加粗细双线（粗线在上，线宽0.8mm），双线上居中打印页眉文字，奇偶页页眉文字均为“哈尔滨工业大学远程教育本（专）科毕业设计(论文）”（五号字打印）。
人工神经网络的最新发展综述摘要:人工神经网络是指模拟人脑神经系统的结构和功能,运用大量的处理部件,由人工方式建立起来的网络系统。
该文首先介绍了神经网络研究动向，然后介绍了近年来几种新型神经网络的基本模型及典型应用,包括模糊神经网络、神经网络与遗传算法的结合、进化神经网络、混沌神经网络和神经网络与小波分析的结合。
最后,根据这几种新型神经网络的特点,展望了它们今后的发展前景。
关键词:模糊神经网络;神经网络与遗传算法的结合;进化神经网络;混沌神经网络；神经网络与小波分析。
The review of the latest developments in artificial neuralnetworksAbstract:Artificial neural network is the system that simulates the human brain’s structure and function, and uses a large number of processing elements, and is manually established by the network system. This paper firstly introduces the research trends of the neural network, and then introduces several new basic models of neural networks and typical applications in recent years, including of fuzzy neural network, the combine of neural network and genetic algorithm, evolutionary neural networks, chaotic neural networks and the combine of neural networks and wavelet analysis. Finally, their future prospects are predicted based on the characteristics of these new neural networks in the paper.Key words: Fuzzy neural network; Neural network and genetic algorithm; Evolutionary neural networks; Chaotic neural networks; Neural networks and wavelet analysis1 引言人工神经网络的研究始于20世纪40年代初。
描述人工智能的主要学派并说明各种学派的特点人工智能（Artificial Intelligence，简称AI）是计算机科学的一个领域，通过研究和开发使计算机可以模拟、仿真和实现人类智能行为的技术和方法。
在人工智能的发展过程中，涌现了多种不同的学派，它们通过不同的方法和理论来研究和实现人工智能。
以下是人工智能的主要学派及其特点。
1. 符号学派（Symbolic AI）：符号学派是人工智能发展的早期学派，强调使用逻辑和符号表示知识，并利用推理和搜索算法进行问题求解。
符号学派注重人工智能的逻辑推导和符号操作能力，其代表性算法包括专家系统和语义网络。
然而，由于复杂问题的表示和推理困难，符号学派受到了知识获取和效率等方面的限制。
2. 连接主义学派（Connectionist AI）：连接主义学派认为人工智能可以通过模拟人脑神经网络的方式来实现。
它采用了一种分布式表示方式，将知识存储在神经元之间的连接权重中。
连接主义学派的一个重要模型是人工神经网络（Artificial Neural Networks），通过训练和学习来提取和处理信息。
这种学派的特点是其能力通过大规模数据的训练和学习不断提高，但对于知识的表示和解释能力相对较弱。
3. 进化学派（Evolutionary AI）：进化学派受启发于生物进化理论，认为通过模拟自然选择和进化的方式来达到人工智能。
该学派利用遗传算法、遗传程序设计等方法进行问题求解和优化，通过不断迭代和演化来优化解决方案。
进化学派的特点是具有自适应性和强大搜索能力，但在处理复杂问题上存在计算复杂度高和时间开销大的问题。
4. 感知学派（Perceptual AI）：感知学派注重模仿人类感知和认知过程，对计算机进行视觉、听觉和语音等感知能力的研究。
该学派的研究重点是利用模式识别、计算机视觉和语音识别等技术实现计算机对外部世界的感知和理解。
感知学派的特点是强调数据驱动和现实世界的应用，但对于高级推理和决策能力的研究较少。
1.Farid F. Abraham 法里德·F·亚伯拉罕(born May 5, 1937) isan American scientist.He has pioneered new methods of using computer modeling in the fields of fracture mechanics, membrane dynamics and phase transformation behavior of matter. He has written two textbooks and over 200 papers published in international journals. He won the Aneesur Rahman Prize in Computational Physics, which is the highest prize given by the American Physical Society.2. Thomas Adams (May 4, 1818 –February 7, 1905),[1] was a 19th-century American scientist and inventor who is regarded as a founder of the chewing gum industry. He eventually joined with well-known chewing gum maker William Wrigley, Jr.4. Daniel John "Dan" Alderson (October 31, 1941 – May 17, 1989) was a scientist at the Jet Propulsion Laboratory in California, and a prominent participant in science fiction fandom. He came from a middle-class family and had diabetes. A high school science fair project on the gravitational fields of non-spherical bodies won him a college scholarship to Caltech and a job at the Jet Propulsion Laboratory, where he wrote the software used to navigate Voyagers 1 and 2.Paul Wheaton is a contemporary permaculture theorist,[1][2] master gardener,[2]software engineer,[1][2][3] and disciple of natural agriculturist Sepp Holzer.[2]Geoff Lawton has called Paul Wheaton "The Duke of Permaculture" for being known as the founder[1][2][4] of websites forums, articles, videos and podcasts suchas which is believed to be the largest website devoted to permaculture.[2]John Polk Allen (born 6 May 1929, Carnegie, Oklahoma)[1] is a systems ecologist and engineer, metallurgist, adventurer and writer.[2] He is best known as the inventor and Director of Research of Biosphere 2, the world's largest laboratory of global ecology, and was the founder of Synergia Ranch. Allen is a proponent of the science of biospherics. Biosphere 2 set a number of world records in closed life system work including degree of sealing tightness, 100% waste recycle and water recycle, and duration of human residence within a closed system (eight people for two years). Allen has also conceived and co-founded nine other projects around the world, pioneering in sustainable co-evolutionary development.Allenby has authored a number of articles and book chapters on his above mentioned interests, and writes a column for the Journal of Industrial Ecology.David Alter is credited with having invented:∙1836 - the electric telegraph, predating the Morse telegraph in 1837.[2][3][4]∙1840 - his electric buggy - the forerunner of the automobile.∙1845 - a patented method to manufacture and purify bromine from salt wells, highly useful in the iron industry and displayed in the World's Fair of 1853(see: Exhibition of the Industry of All Nations in New York City).∙1854 - spectrum analysis, the idea that every element has its own emission spectrum: a breakthrough development in spectroscopy. The published articlewas:On Certain Physical Properties of Light Produced by the Combustion of Different Metals in an Electric Spark Refracted by a Prism.[5] He included a chart of the colored lines or bands of twelve metals and paved the way by showing the spectral lines of brass corresponded to copper and zinc.∙1855 - an expansion of spectrum analysis to include the optical properties of gas.[6] These discoveries were later implemented and included by GustavKirchhoff and Robert Bunsen in the Three Laws of Spectroscopy.∙1858 - a patented method to extract oil from coal and shale, along with a partner Samuel Hill. Their invention sped manufacturing, but was replaced by technology in a few years.∙An electric clock.∙ A short range type of telephone - forerunner of the Graham Bell telephone.Jerome III "Jay" Apt, Ph.D. (born April 28, 1949 in Massachusetts) is an American astronaut and professor at Carnegie Mellon University. Before he became an astronaut, Apt was a physicist who worked on the Venus space probe project, andsystem from ground-based observatories.Frances Hamilton Arnold (born 25 July 1956) is an internationally recognized American scientist and engineer. She pioneered methods of directed evolution to create useful biological systems, including enzymes, metabolic pathways,genetic regulatory circuits, and organisms. She is the Dick and Barbara Dickinson Professor of Chemical Engineering,Bioengineering, and Biochemistry at the California Institute of Technology, where she studies evolution and its applications in science, medicine, chemicals and energy. She earned her B.S. in Mechanical and Aerospace Engineering from Princeton University in 1979 and her Ph.D. in Chemical Engineering from the University of California, Berkeley. There, she did her postdoctoral work in biophysical chemistry before coming to Caltech in 1986.Arnold's Caltech research is in green chemistry and alternative energy, including the development of highly active enzymes (cellulolytic and biosynthetic enzymes) and microorganisms to convert renewable biomass to fuels and chemicals. She is co-inventor on numerous patents and co-founded Gevo, Inc. in 2005Joseph Brant Arseneau (born September 3, 1967) Arseneau's primary area of technical expertise is in biologically-inspired systems called computational intelligence,[4] which include: neural networks,[5]genetic algorithms, fuzzy logic, swarm intelligence, and intelligent agents. His interest in computational intelligence began with his research into the application of neural networks to software reengineering;[6] that research went on to form the foundation to a patent being awarded to Raymond Obin and Brian Reynolds for a commercial reenginering process.Siva Subrahmanyam Banda (born 1951)Dr. Banda holds two patents - one for a control mechanism for unknown systems which can be applied to an arbitrary electromechanical system without prior knowledge of itsCharles E. Bartley (1921–1996) was an American scientist, known for developing the first elastomeric solid rocket propellant formula, at the Jet Propulsion Laboratory (now part of NASA) in Pasadena, California in the late 1940s.Robert Nason Beck (26 March 1928 in San Angelo, Texas– 6 August 2008 in Chicago, Illinois) was an American scientist and a pioneer in the field of nuclear medicine. Part of a University of Chicago team, he was the first to propose, in 1961, the use of the radioisotope technetium-99m to detect disease using Positron Emission T omography, a technique that is used an estimated 20 million times a year throughout the world.[1] Beck also helped develop collimators for sharpening the images produced by gamma-ray scanners, and was referred to as 'Mr. Collimator' by colleagues.James (Jim) F. Bell IIIBell is an active planetary scientist and has been involved in many NASA robotic space exploration missions. As a professional scientist, he has published over 30 first-authored and 140 co-authored scientific research papers and over 400 short abstracts and conference presentations. Bell has also written and edited several books about Mars and the Moon. He is active in educating the public about space exploration. He is a frequent contributor to popular astronomy and science magazines, has made a number of television appearances on major network and cable channels, and gives free public lectures.May Roberta Berenbaum (born 1953) is an American entomologist whose research focuses on the chemical interactions between herbivorous insects and their host-plants, and the implications of these interactions on the organization of natural communities and the evolution of species.Since 1980, Berenbaum has been a member of the faculty of the department of entomology at the University of Illinois in Urbana-Champaign and has served as head of the department since 1992.[2] In addition to her research, she is devoted to teaching and to fostering scientific literacy. In 1996, she was elected a Fellow of the American Academy of Arts and Sciences.[3] she is the recipient of the 1996 Entomological Society of America North Central Branch Distinguished Teaching Award and has authored numerous magazine articles, as well as three books about insects for the general public. She has also gained some measure of fame as the organizer of the Insect Fear Film Festival at the University of Illinois, an annual celebration of Hollywood's entomological excesses.Harvey Bialy (born 1945, New York City) is an American molecular biologist and AIDS denialist. He was one of the signatories to a letter to the editor by a group of AIDS sceptics calling themselves the Group for the Scientific Reappraisal of the HIV-AIDS Hypothesis.[1] He was also a member of the controversial and heavily criticized South African Presidential AIDS Advisory Panel convened by Thabo Mbeki in 2000.James Bloodworth, Jr.He employed histochemistry and ultrastructural studies to assess lesions of the pancreatic islet cells and blood vessels in patients with diabetes mellitus.Mark BosloughAn expert on planetary impacts and global catastrophes, Boslough’s work o n airbursts challenged the conventional view of asteroid collision risk and is now widely acceptedKarel Jan BossartAs a result Bossart's achievements include:∙Launch of first communications satellite∙Launch of first US orbital manned missions∙Launch of Mariner probes to Mars and Venus∙Launch of Pioneer 10 and Pioneer 11 to Jupiter and Saturn.William Clouser BoydRobert Stephen BoyerHe and J Strother Moore invented the Boyer–Moore string search algorithm, a particularly efficient string searching algorithm, in 1977. He and Moore also collaborated on the Boyer–Moore automated theorem prover, Nqthm, in 1992.[1] Following this, he worked with Moore, and Matt Kaufmann on another theorem prover called ACL2.。
吴恩达2篇吴恩达是计算机科学家、人工智能专家，也是深度学习领域的奠基人之一。
在他的职业生涯中，他做出了许多重要的贡献，推动了人工智能的发展。
本篇文章将分别介绍吴恩达的两篇经典论文，分别是《Unsupervised Feature Learning and Deep Learning》和《Deep Learning of Representations for Unsupervised and Transfer Learning》。
第一篇论文《Unsupervised Feature Learning and Deep Learning》是吴恩达在2012年发表的。
这篇论文主要介绍了无监督特征学习和深度学习的方法，并提出了一种新的算法Deep Belief Networks（DBNs）。
DBNs是一种无监督学习的神经网络，通过多层次的学习来学习特征表示。
本篇论文首先介绍了传统的无监督特征学习方法，如自编码器和K-means等。
然后，吴恩达介绍了DBNs的工作原理和训练过程。
DBNs 由多个Restricted Boltzmann Machines（RBMs）组成，每个RBMs都学习到一组特征，并将其传递给上一层。
通过逐层的学习，DBNs能够学习到更高级别的特征表示，从而提取数据中的有用信息。
吴恩达还介绍了DBNs在图像分类和语音识别等任务中的应用，并展示了其与传统方法相比的优势。
通过无监督学习的特征表示，DBNs能够更好地处理大规模数据和复杂问题，并取得更好的性能。
第二篇论文《Deep Learning of Representations for Unsupervised and Transfer Learning》是吴恩达在2013年发表的。
这篇论文进一步探讨了深度学习的特征表示和迁移学习的问题。
吴恩达认为，深度学习的关键在于学习到具有高度可区分性和泛化能力的特征表示。
本篇论文首先介绍了深度学习中的监督学习和无监督学习。
Evolutionary Design of Neural Networksfor Classification and Regression1Miguel Rocha ,Paulo Cortez†,Jos´e Nevesrm´a tica,University of Minho,Portugal†Dep.Sistemas de Informac¸˜a o,University of Minho,Portugal E-mail:mrocha@di.uminho.pt pcortez@dsi.uminho.pt jneves@di.uminho.ptAbstractThe Multilayer Perceptrons(MLPs)are the mostpopular class of Neural Networks.When applyingMLPs,the search for the ideal architecture is a crucialtask,since it should should be complex enough tolearn the input/output mapping,without overfittingthe training data.Under this context,the use ofEvolutionary Computation makes a promising globalsearch approach for model selection.On the otherhand,ensembles(combinations of models)have beenboosting the performance of several Machine Learning(ML)algorithms.In this work,a novel evolutionarytechnique for MLP design is presented,being alsoused an ensemble based approach.A set of real worldclassification and regression tasks was used to test thisstrategy,comparing it with a heuristic model selection,as well as with other ML algorithms.The results favourthe evolutionary MLP ensemble method.Keywords:Supervised Machine Learning,MultilayerPerceptrons,Evolutionary Algorithms,Ensembles.1IntroductionNeural Networks(NNs)are important Machine Learn-ing(ML)techniques,denoting a set of connectionistmodels inspired in the behavior of the human brain.Inparticular,the Multilayer Perceptron(MLP)is a populararchitecture,where neurons are grouped in layers andonly forward connections exist.This provides a pow-erful base-learner,capable of nonlinear mappings[1].When compared to other ML methods,MLPs are knownto behave well in terms of predictive knowledge[2],and there has also been research in terms of explanatoryknowledge(e.g.extracting rules from MLPs)[3].However,one of the major issues when applyingMLPs is the topology(i.e.connectivity)design.This isa complex and crucial task,with a strong impact in per-formance(a small network may provide poor learningtested in classification and regression tasks,using both single and ensemble based models.Finally,results will be compared with a heuristic NN selection procedure,as well with other ML methods.2Materials and Methods2.1Data SetsEight classification and eight regression data sets were selected from the UCI ML repository[9],and its main features are listed in Table1,namely the number of nu-meric(Nu),binary(Bi)and nominal(No,i.e.discrete with3or more labels)input attributes,as well as the number of examples(Ex)and classes(Cl).The regres-sion tasks are identified by the symbol in the Cl col-umn(last eight rows).Table1.A summary of the data sets used.Balance4006253Bupa6003452Car00617284Cmc53114733Dermatology34003666Ionosphere34003512Sonar60001042Yeast7101484102.2Neural NetworksBefore feeding the MLPs,the data was preprocessed: a1-of-C encoding(one binary variable per class)was applied to the nominal attributes and all input values were rescaled within the range[−1,1].For example,the safety attribute from the task car was encoded according to:low→1-1-1,med→-11-1and high→-1-11. Regarding the outputs,the discrete variables were nor-malized within the range[0,1](using also a1-of-C en-coding for the nominal attributes).Therefore,the pre-dicted class is given by the nearest class value to the node’s output,if one single node is used(binary vari-able),otherwise the node with the highest output value is considered.On the other hand,regression problems will be modeled by one real-valued output,which di-rectly represents the dependent target variable.The MLPs used make use of biases and sigmoid ac-tivation functions,with one hidden layer,containing a variable number of nodes.A different approach was fol-lowed for the regression tasks,since outputs may lie out of the logistic output range([0,1]).In this case,the lo-gistic function was applied on hidden nodes,while the output ones used shortcut connections and linear func-tions,to scale the range of the outputs(Figure1).This solution avoids the need offiltering procedures,which may give rise to information loss and has been success-fully adopted in other regression applications,such as Time Series Forecasting[10].xx12Fig.1.A2−2−1MLP topology with bias and shortcuts.The initial weights will be randomly set within the range[−1,1].Then,the RPROP algorithm[11]is se-lected for training,due to its faster convergence and sta-bility,being stopped after a maximum of200epochs or when the error slope is approaching zero.Two distinct accuracy measures were adopted:the Percentage of Correctly Classified Examples(PCCE), used in classification tasks;and the Normalized Root Mean Squared Error(NRMSE),applied in the regression ones.These metrics are given by the equations:P CCE=P N i=11,if(T i=P i)P N i=1(T i−P i)2P N i=1T i2.3Evolutionary Neural NetworkIn this work,an Evolutionary Algorithm with a direct representation is embraced,where the genotype is the whole MLP.Each individual of the initial population is set by choosing a random number of hidden nodes(be-tween0and10).Then,each possible connection is set with probability of50%.New individuals are bred by structural mutation,which works by adding or deleting a random number(from1to5)of nodes or connections. The population size was set to20,being the selection done by converting thefitness value(the error computed over a validation set)into its ranking,and then applying a roulette wheel scheme,being used a substitution rate of 50%.Finally,the ENN is stopped after20generations. This scalable ENN is able to search through any kind of MLP connectivity,ranging from linear models to com-plex nonlinear MLPs.The ENN Ensemble(ENNE)will be built using the best20individuals(with lower vali-dation error)obtained during the evolutionary process, being the output computed as the average of the MLPs.3ResultsThe NN/EC experiments were conducted using a soft-ware package developed in Java by the authors.The other techniques were computed using WEKA ML soft-ware(with its default parameters)[12]:•J48–a classification decision tree based on theC4.5algorithm;•M5P–a regression decision tree(M5algorithm);•IB5–a5-Nearest Neigboor;•KStar–an instance based algorithm;and•SVM–a Support Vector Machine.For each model,10runs of a5-fold cross-validation pro-cess[13](stratified in the classification tasks)were exe-cuted.This means that in each of these50experiments, 80%of the data is used for learning and20%for testing. Regarding the MLP based approaches,the learning data was divided into training(50%of the original dataset) and validation sets(30%).Tables2and3show the av-erage errors of the10runs for each learning model and classification/regression task.In both tables,the last row averages the global behaviour of each technique. First,the classification results will be analyzed.The NN based approaches(last four columns)are competive when compared with the other ML algorithms.The few exceptions are the dermatology and sonar tasks,where the SVM and KStar get the best results.As expected,the ENN outperforms the HNN,with a1%increase in perfor-mance.Furthermore,the ensemble approaches(HNNE and ENNE)obtain better results when compared with the single based methods(1.4%improvement in both cases). Indeed,the ENNE reveals the best overall behaviour,out-performing all other algorithms in5of the8tasks.A similar scenario occurs in the regression tasks.In general,the NN methods are better than the other ML al-gorithms,although the M5P outperforms the HNN and HNNE approaches.When compared with the ENN,the HNN is outperformed by a wider difference(2.5%).As before,the ensembles behave better,although the impact is higher with the ENNE(1.5%improvement,being the best method in5tasks)than with the HNNE(0.7%im-provement).4ConclusionsThe surge of bio-inspired techniques,such as Multi-layer Perceptrons(MLPs)and Evolutionary Computa-tion(EC),has created new exciting possibilities for the field of Machine Learning(ML).Considering ensembles of learning models to improve its accuracy has also been a focus of attention by the research community.In this work,an evolutionary approach to MLP topology design was presented,considering single and ensemble combi-nations,being tested in supervised learning tasks(e.g., classification and regression).The results obtained confirm than Evolutionary Neu-ral Network(ENN)approach outperforms a heuristic trial-and-error MLP design procedure(HNN),as well as other ML algorithms(e.g.K-Nearest Neigboor).How-ever,this improvement in performance has the handicap of increasing the computational complexity(the HNN re-quires only a tenth of the ENN computational effort). On the other hand,the use of the EC population struc-ture to construct ensembles is a recent researchfield.In-deed,the proposed ENN Ensemble(ENNE),based on the average of the outputs from the best EC individuals(or MLPs),is competitive.The ENNE has the advantage of presenting the best overall performance while requiring the same computational effort,when compared with the single based ENN.In future work,it is intended to explore similar ap-proaches with different neural architectures(e.g.,Recur-rent Neural Networks).Moreover,more elaborated en-sembles should be considered,by designingfitness func-tions which reward specialization[14].References[1]Haykin,S.(1999)Neural Networks-A Compreen-sive Foundation.2nd ed.Prentice-Hall[2]Quinlan,J.R.(1994)Comparing Connectionist andSymbolic Learning Methods.In:Hanson,S.et al.(eds.)Computation Learning:Theory and Natural Learning Systems,MIT Press,pp.445-456Table2.The classification results(P CCE values,in%).Balance78.187.688.387.794.895.895.796.5 Bupa64.860.765.958.068.469.068.569.7 Car91.392.387.193.597.498.298.398.7 Cmc51.247.249.648.450.654.252.155.9 Dermatology95.796.694.597.494.995.595.996.8 Ionosphere89.484.684.087.988.989.692.392.6 Sonar72.580.585.276.779.979.080.981.2 Yeast56.057.153.156.658.259.359.960.2 Table3.The regression results(NRMSE values,in%).Abalone24.325.324.825.023.223.223.223.2 Auto-mpg11.815.114.613.514.213.012.212.2 Autos13.321.421.613.814.616.114.214.9 Breast-cancer40.840.843.644.042.640.647.440.2 Heart-disease21.221.025.521.521.921.722.221.5 Housing18.422.218.022.618.417.316.616.0 Servo50.460.467.670.560.845.049.639.4 Wpbc73.273.698.271.775.474.480.371.7[3]Setiono,R.(2003)Techniques for ExtractingClassification and Regression Rules from Artifi-cial Neural Networks.In D.Fogel and C.Robin-son,(eds.),Computational Intelligence:The Ex-perts Speak,IEEE Press/Wiley,pp99–114 [4]Thimm,G.and Fiesler,E.(1995)Evaluating prun-ing methods.In:Proc.of the Int.Symp.on Artifi-cial Neural Networks,pp20–25[5]Kwok,T.and Yeung,D.(1997)Constructive al-gorithms for structure learning in feedforward neu-ral networks for regression problems problems:A survey.IEEE Transactions on Neural Networks, 8(3):630–645[6]Yao,X.(1999)Evolving Artificial Neural Net-works.In:Proc.of the IEEE,87(9):1423-1447 [7]Dietterich,T.(1997)Machine Learning Research:Four Current Directions.AI Magazine,18(4):97–136[8]Rocha,M.,Cortez,P.and Neves,J.Ensemblesof Artificial Neural Networks with Heterogeneous Topologies.In:Proc.of the4th Symposium on En-gineering of Intelligent Systems(EIS2004).ICSC Academic Press[9]Blake,C.and Merz,C.(1998)UCI Repository ofMachine Learning Databases,University of Cali-fornia[10]Cortez,P.,Rocha,M.and Neves,J.(2001)Evolv-ing Time Series Forecasting Neural Network Mod-els.In:Proc.of the3rd Int.Symposium on Adap-tive Systems:Evolutionary Computation and Prob-abilistic Graphical Models(ISAS2001),pp.84–91 [11]Riedmiller,M.(1994)Supervised Learning inMultilayer Perceptrons-from Backpropagation to Adaptive Learning puter Stan-dards and Interfaces16[12]Witten,I.and Frank,E.(2000)Data Mining:Prac-tical Machine Learning Tools and Techniques with Java Implementations.Morgan Kaufmann [13]Kohavi,R.(1995)A Study of Cross-Validation andBootstrap for Accuracy Estimation and Model Se-lection.In:Proc.of the Int.Joint Conference on Artificial Intelligence(IJCAI)[14]Liu,Y.,Yao,X.and Higuchi,T.(2000)Evolution-ary Ensembles with Negative Correlation Learning.IEEE Transactions on Evolutionary Computation, 4(4):380–387。
