Ӧо
APPLICATION RESEARCH OF COMPUTERS
2000Vol.17No.2P.34-37



ȡо
ﳿ־Ǭ
ժ Ҫ Ѿںܹ㷺õӦãѵõе֪ʶ⣬类Ϊһ͵ē䔽ṹгȡʾ֪ʶǽһЧֶΡһЩдԵȡ㷨һЩδоص㡣
ؼ  ȡ ѧϰ
1 
缼Դ80˽ϴķչǿķԴϴϢ걸ԼݵȡôͳѧϰԴﵽЧǣѵõģǵ͵ē䔽ṹȡ֪ʶױ⣬ʹھ֪ʶȷӦܵ˺ܴơΪ˳ַ缼سȡе֪ʶѾΪеһǳҪо⡣
ȡģе֪ʶҪо֧һо򣻶огȡ򡣱Ҫۺߣڷеĵ͹ȡ㷨ϣȡδоչһЩ
2 ȡоҪ
缼ʮѾɹӦںܶ򣬵ܵɡҪԭͨѵѧϰ֪ʶֵʽ洢У֪ʶʾķֲʽԣʹԶ˽⣬Ҳ޷ȷ̵Ľ͡ڶ֪ʶʾĿѧϰ㷨ӵҪ۱׼֮һ[1]˹ȡѧϰоҪ塣гȡĿǽ֪ʶһķʽȷرһ棬Թʽʾȡ֪ʶԷʹߺ̣һ棬ͨȡĹûԷһЩǰԵҪϵӶûоߴ⣬˿Ϊѵõ֪ʶȡΪһڸֲͬĹѧϰĹߣЧԺȷҲѾõ֤ʵ
ΪȡҪ壬ⷽĹܵԽԽĹעIEEE Transaction on Neural Networks1998רųһרTickleƪȷгȡǵǰ缱[2]˵˶ԹȡоĿǰѾΪһҪо
3 ȡ㷨
ȡо翪ʼ80ĩGallant1988[3]һIF-THEN۵רϵͳԴԺ󣬺ܶо߶˹ȡоȡ˴ĳɹݳȡ㷨˼ĲͬԷֳɻڽṹ㷨ͻܷ㷨ࡣݳȡ㷨õṹͬԷֳṹ㷨ͨ͹ȡ㷨ݳȡĹʽֿͬɷΪȡͨIF-THEN㷨ȡģ㷨ȡ㷨Լȡ㷨ȡڱУǴӹȡ㷨˼ֱԻڽṹĺͻܷĵ㷨нܺͷ
3.1 ڽṹĹȡ㷨
ڽṹȡ㷨ѹȡһ̡˼ǰѵõṹӳɶӦĹͼ1ͱʾһ򵥵Ĺȡ㷨

ͼ1򵥵Ĺȡ
̵ļ㸴ӶȺָ֮Ĺϵܶʱϱը⡣ˣ㷨һü֦ȷеԽͼ㸴ӶȡȽϵ͵㷨¼֣
(1)Gallant[3]
о80ĩ Gallantһ͵רϵͳһ򵥵Ĺȡ㷨ڽ͸רϵͳ
㷨ȡΪĳ(Case)óۡˣ˼Ǵӵǰ֪ϢѡһЧزý۵СϢϡҲ˵δ֪ȡֵΪ٣ֻҪСϢϵȡֵҪͿԵóýۡ
ϸ˵㷨һϵĹȡ㷨ҪĿĲǽйȡϣһ֪۵һĽͣ㷨򵥣ֻС͵硣
(2) Subset㷨[4]
FuSubset 㷨ȡ㷨֮һĹȡ㷨ĻϷչġ
Subset㷨ü򵥵Ĺ㷨ȡͨIF-THENSubsetɵļϵݼһӼ㷨ɵɵĳʼSubsetϿʼԣÿԱܷЧر֤ƫֵԵĻùSubsetĴСͬķԲֱпܵSubsetɾĹ
Subset㷨ûвýͼ㸴ԵļֻʺϴС͵⡣SaitoNakano[5]KT㷨޶ǰĿķSubset㷨˸ĽԷΪԡ(Pos-atts)͡ԡ(Neg-atts)ֱȻĸԵҲ˳ȡ޷ԭоȷ
(3) MOFN㷨[4]
MOFN㷨ǻTowell ShavlikKBANN(Knowledge based networks)һֹȡ㷨(ͼ2ʾ)

ͼ2ȡMOFN
ȡĹʽΪM-of-NĹ
if (M of N antecedents are true) then ...........
MOFN㷨Ĺؼȼ(Equivalence class)ĸ㷨ʹñ׼ľ㷨ȨֵӽӺϲȼ࣬ÿȼȨֵɸƽֵȻȥЩԽӰ첻ĵȼ࣬ڲȨֵǰ¶½ѵֱӸṹȨֵȡ򣬲򻯡
ʹһ㷨ѵõеȨɢطֲȨֵռУԶдMOFN㷨KBANN硣Ϊչ÷ΧCraven1993Ȩ(Soft weight-sharing)ķѵ[1]ѵйȨֵ࣬Ȼٶø÷ѵõʹMOFN㷨ЧسȡMOFN
MOFN㷨ȡĹ򼯱Ƚϼ׶ֹʽҲ˲Ĺ⣬ڲ˵ȼķӽ˾࣬Ҳʹȡʱ仨Ѵ١
MOFN㷨Ҫȱڣȣ㷨ҪеÿԪļֵΪ˫ֵģʽͶԼѡĳЩ޶ӶĹܡΣ㷨ҪԵȡֵΪɢֵʹѧϰϵͳڴԷԵõӣ㷨ҪÿԪʾһΨһĸͨķֲʽ֪ʶʾнϴĳͻ÷Χ
(4) RX(Rule-extraction)㷨[6]
㷨Setionoһڱ׼ǰͨ͹ȡ㷨㷨ȡͨIF-THEN
㷨ȡ㷨ĸĽ(1)ԪļֵͨɢҲ˵һɢֵʾÿԪļֵ(2)öԪٻֵļ
÷ÿԪֱӸӵԪɢļֵӦĹ򡣶ԪȿӵԪĸ֮ԪĿ׼ֱӶӦ򡣷򣬽һ磬ĸΪԪɢֵĿĸΪԪԪĸѵٵñ㷨ȡȻRX㷨ԵݹĴʽйȡ
⣬ų[12]һRX㷨緢ַ㷨صǲŴ㷨(GA)ṹм֦
ڴ󲿷ֻȡ㷨Ϊ˼ʱòƹǰĸ㷨ȡԪٹķȻ㷨ѵʱ䣬ԸΪȷ
㷨ڱ׼ǰ硣ΪЧز㷨ҪȽм֦ȥԽӰ첻ӡڹȡпܵݹɶ磬˸㷨ʱ俪ܴڶԼֵɢǰ㷨RX㷨ԼֵķΧƣ㷨ҪΪɢԣҪҪɢ[7]
3.2 ܷĹȡ㷨
ڽṹ㷨ܷͬȡ㷨ѹȡ̿һṹḶ́ǰ翴һڹȡвǾṹ
(1) RF(Rule from facts)㷨RN(Rule from networks)㷨[8]
SaitoNakano㷨ڳȡDNF(Disjuctive Normal Form)ʽĹRF㷨ʹʽռ֦ʵֱӳȡRN㷨ѵõгȡRN㷨ȸһʵ(Positive fact)ͨԵȡֵΧһȻαʵ(Negative fact)Сùֱù򲻰κηΪֹ
㷨ȡ㷨֮һ翴һ壬ľṹҪͨʵȡ
(2) ѧϰĹȡ㷨
ѧϰȡ㷨ȡΪһѧϰ⡣Craven Shavlik ݸ˼TREPAN㷨ȡʽĹ[9,10]Craven Ķ壬ȡǸһѵõԼڶѵݣһĸͬķʽʾл֡
ؼڰѹȡһݺѧϰ̡㷨Ļ˼ṩشʣһʵ(Example Oracle)زʵʵ򡣼򵥵˵һʵֱӽӦһĹҲĵһڴ˻ϣɾһԾͿԲļɾǰԽ࣬ķԽǿǵʵķҲԽҪԹвԡһӵ(Subset Oracle)ѧϰģ͵ĲͬͬҪܾǸѾѵõѧϰģͲԸļǷȷÿζԸʵݿһвԣһȷĹ򼯣󽫸ùǰĿٵ⡣
㷨صһͨ͵Ĺȡ㷨ضѧϰģ͵ƣԴӸԼѧϰϵͳгȡĹԷ϶㷨ļ㸴Ӷȱ㷨Ҫ͡㷨ҪȱڼĲʽڴɢԣʺԵĴҪԣҪڹȡ֮ǰԽɢ
(3) VI(Validity-Interval Analysis)Ĺȡ㷨[11]
ThrunһֻЧ(Validity-Interval Analysis)Ĺȡ㷨㷨ṹһںУֻӰ죬㡣㷨ȡĹ繫ʽ(1)

THEN xiʾĸΪ桡(1)
xiʾĳԣʾԸԵȡֵռһ֡㷨ѵʵвĹΪӹVIԸùвԣݽȡֵռĻֱΪֹѭµӹĽֱĳ׼
㷨ṹƣĹֱӴӳ䵽мԪٵӰ죬ڸֶ磬ҶԵȡֵûơȡĹ򾫶ȱȽϸߣʽʾǰʹùĿԺܵ͡Ҹ㷨ļ㿪ʮְʵӦǲеģ˸㷨Ҫ;ǽйԣڹȡ
(4) Benitez㷨
BenitezCastroRequena[11]׼ǰѧϰϵͳͻģѧϰϵͳǵȼϵͳ֤ʹòΪмԪǰһӦFAS(fuzzy Additive System)еÿ(ԪjԪk)ֱӶӦµһģ繫ʽ(2)
(2)
ARϵģȺΪļxiiԪֵwijiԪ͵jмԪ֮ȨֵjΪjмԪƫֵykkԪֵjkjмԪ͵kԪ֮ȨֵǰĻ㷨Լ֤ʽȷԡȻø÷Լ򵥵سȡ򣬵ĿԺܲʵӦ;
4 һо
Ϊһֻȡṹ֪ʶֶΣȡ㷨ɻ˽ϴĳɹ͸һѴ֪ʶḻēںДԿסеȡ㷨Ӧкܶ಻֮Щ֮ĸĽѾΪµоĿꡣһоҪ·棺
(1) ԵĴ
󲿷ֹȡ㷨ṹӦ÷ΧҪԭֻܴɢԣΪԭڼ򵥵ĹԵıԿʵĹʽʹñȽϼ򵥵ģԣûع(Regression trees)ΪȡĹıʽ[4]ҲԲñȽϺõľ㷨ڹȡиҪɢѵʱɢ
(2) 㷨Чʵ
ȡ㷨ļ㸴ӶȿǸüһӦõһҪء㷨Чʣͼ㸴ӶҲǽص㡣ԴпǣһǿԲøּ֦㷨ṹĸԣǿͨѧϰռ䣬øʽͼ㸴Ӷȡ
(3) ʾʽо
е㷨ڳȡĹĿԼʾĸԷҲдߡԿʽĹʾʽôȶȵĹʾʽ͹Ŀ͹ʾĸԣ⣬ҲԿǶԳȡĹ򼯲ü֦㷨[1]߹ĿԺ͹ķ
(4) 缯гȡ
Ŀǰ缯(Ensemble)[13]𽥳Ϊһоȵ㡣üԴڻѧϰĿǰŵBoosting[14]Ȼ缯Ԥ⾫ȣԱȵһͣоδ缯гȡĹΪȡһҪо
5 С
ȱ֪ʶĽһֱ谭缼㷺ӦõҪ֮һгȡо˿ķʽѧϰ֪ʶ⣬֪ʶֵоרϵͳе֪ʶȡоԼоں缼ͷѧϰовοֵ
еĸȡ㷨Ųͬȱݣ㷨ļ㸴ԽϸߣĿԽϲȡţŹȡоĲ뽫ʹ缼õΪ㷺Ӧá
ùȻѧ
ﳿϾѧ¼صʵ Ͼ 210093
־Ͼѧ¼صʵ Ͼ 210093
ǬϾѧ¼صʵ Ͼ 210093
ο
1Craven M, Shavlik J. Learning Symbolic Rules Using Artificial Neural Networks. In: Proceedings of the Tenth International Conference on Machine Learning, Amherst, MA, Morgan Kaufmann, 1993, 7380
2Alan B.Tickle,Robert Andrews, The Truth Will Come to Light:Directions and Challenges in Extracting the Knowledge Embedded Within Trained Artificial Neural Networks. IEEE TRANSACTIONS ON NEURAL NETWORKS,Vol 9,NO 6 NOVEMBER 1998 10571068
3Stephen I.Gallant, Connectionist Expert Systems Communication of the ACM. Vol 31,NO 2 1988 152169
4Towell G.G Shablik,J.W,(1993), Extracting Refined Rules From Knowledge-Based Neural Networks Machine Learning Vol.13 ,1, 1993
5Fu,L.M (1991) Rule Learning by Searching on Adapted Nets. In Porceeding of the Ninth National Conference on Artificial Intelligence,pp 590595, Anaheim,CA:AAAI Press
6Setiono R. Extracting Rules from Neural Networks by Pruning and Hidden-Unit Splitting. Neural Computation, 1997, Vol.9, No.1, 205225
7Craven M, Shavlik J. Extracting Tree-Structured Representations of Trained Networks. In: Touretzky D, Mozer M, Hasselmo M, eds., Advances in Neural Information Processing Systems (Volume 8), The MIT Press, Cambridge, MA, 1996, 2430
8Saito K, Nakano R. Rule Extraction from Facts and Neural Networks. In: Proceedings of the International Neural Network Conference (INNC-90), Paris, France, Kluwer Academic Publishers, Netherlands, 1990, 379382
9Craven M, Shavlik J. Using Sampling and Queries to Extract Rules from Trained Neural Networks. In: Proceedings of the Eleventh International Conference on Machine Learning, New Brunswick, NJ, Morgan Kaufmann, 1994, 3745
10Thrun S Extracting Rules from Artificial Neural Networks with Distributed Representations. In: Tesauro G, Touretzky D, Leen T, eds., Advances in Neural Information Processing Systems (Volume 7), The MIT Press, Cambridge, MA, 1995
11Benitez J, Castro J, Requena I. Are Artificial Neural Networks Black Boxes? IEEE Transactions on Neural Networks, 1997, Vol.8, No.5,.11561164.
12ųԵ. 緢ַ. ѧ, 1999, Vol. 22, No.1, 108112
13Igelnik B,Pao Y, The Ensemble Approach to Neural-Network Learning and Generalization IEEE Transactions on Neural Networks ,1999,10(1):1930
14Schapire R. The Strength of Weak Learnability. Machine Learning, 1990, 5: 197227
ոڣ199996
