Multi-Criteria-based Active Learning for Named Entity Recognition

Size: px
Start display at page:

Download "Multi-Criteria-based Active Learning for Named Entity Recognition"

Transcription

1 Mult-Crtera-based Actve Learnng for Named Entty Recognton Dan Shen 1 Je Zhang Jan Su Guodong Zhou Chew-Lm Tan Insttute for Infocomm Technology 21 Heng Mu Keng Terrace Sngapore Department of Computer Scence Natonal Unversty of Sngapore 3 Scence Drve 2, Sngapore {shendan,zhange,suan,zhougd}@2r.a-star.edu.sg {shendan,zhange,tancl}@comp.nus.edu.sg Abstract In ths paper, we propose a mult-crterabased actve learnng approach and effectvely apply t to named entty recognton. Actve learnng targets to mnmze the human annotaton efforts by selectng examples for labelng. To maxmze the contrbuton of the selected examples, we consder the multple crtera: nformatveness, representatveness and dversty and propose measures to quantfy them. More comprehensvely, we ncorporate all the crtera usng two selecton strateges, both of whch result n less labelng cost than sngle-crteron-based method. The results of the named entty recognton n both MUC-6 and GENIA show that the labelng cost can be reduced by at least 80% wthout degradng the performance. 1 Introducton In the machne learnng approaches of natural la n- guage processng (NLP), models are generally traned on large annotated corpus. However, annotatng such corpus s expensve and tmeconsumng, whch makes t dffcult to adapt an exstng model to a new doman. In order to overcome ths dffculty, actve learnng (sample sele c- ton) has been studed n more and more NLP applcatons such as POS taggng (Engelson and Dagan 1999), nformaton extracton (Thompson et al. 1999), text classfcaton (Lews and Catlett 1994; McCallum and Ngam 1998; Schohn and Cohn 2000; Tong and Koller 2000; Brnker 2003), statstcal parsng (Thompson et al. 1999; Tang et al. 2002; Steedman et al. 2003), noun phrase chunkng (Nga and Yarowsky 2000), etc. Actve learnng s based on the assumpton that a small number of annotated examples and a large number of unannotated examples are avalable. Ths assumpton s vald n most NLP tasks. Dfferent from supervsed learnng n whch the entre corpus are labeled manually, actve learnng s to select the most useful example for labelng and add the labeled example to tranng set to retran model. Ths procedure s repeated untl the model acheves a certan level of performance. Practcally, a batch of examples are selected at a tme, called batchedbased sample selecton (Lews and Catlett 1994) snce t s tme consumng to retran the model f only one new example s added to the tranng set. Many exstng work n the area focus on two approaches: certanty-based methods (Thompson et al. 1999; Tang et al. 2002; Schohn and Cohn 2000; Tong and Koller 2000; Brnker 2003) and commttee-based methods (McCallum and Ngam 1998; Engelson and Dagan 1999; Nga and Yarowsky 2000) to select the most nformatve examples for whch the current model are most uncertan. Beng the frst pece of work on actve learnng for name entty recognton (NER) task, we target to mnmze the human annotaton efforts yet stll reachng the same level of performance as a supervsed learnng approach. For ths purpose, we make a more comprehensve consderaton on the contrbuton of ndvdual examples, and more mportantly maxmzng the contrbuton of a batch based on three crtera: nformatveness, representatveness and dversty. Frst, we propose three scorng functons to quantfy the nformatveness of an example, whch can be used to select the most uncertan examples. Second, the representatveness measure s further proposed to choose the examples representng the maorty. Thrd, we propose two dversty consderatons (global and local) to avod repetton among the examples of a batch. Fnally, two combnaton strateges wth the above three crtera are proposed to reach the maxmum effectveness on actve learnng for NER. 1 Current address of the frst author: Unverstät des Saarlandes, Computatonal Lngustcs Dept., Saarbrücken, Germany dshen@col.un-sb.de

2 We buld our NER model usng Support Vector Machnes (SVM). The experment shows that our actve learnng methods acheve a promsng result n ths NER task. The results n both MUC- 6 and GENIA show that the amount of the labeled tranng data can be reduced by at least 80% wthout degradng the qualty of the named entty recognzer. The contrbutons not only come from the above measures, but also the two sample selecton strateges whch effectvely ncorporate nformatveness, representatveness and dversty crtera. To our knowledge, t s the frst work on consderng the three crtera all together for actve learnng. Furthermore, such measures and strateges can be easly adapted to other actve learnng tasks as well. 2 Mult-crtera for NER Actve Learnng Support Vector Machnes (SVM) s a powerful machne learnng method, whch has been appled successfully n NER tasks, such as (Kazama et al. 2002; Lee et al. 2003). In ths paper, we apply actve learnng methods to a smple and effectve SVM model to recognze one class of names at a tme, such as proten names, person names, etc. In NER, SVM s to classfy a word nto postve class 1 ndcatng that the word s a part of an entty, or negatve class -1 ndcatng that the word s not a part of an entty. Each word n SVM s represented as a hgh-dmensonal feature vector ncludng surface word nformaton, orthographc features, POS feature and semantc trgger features (Shen et al. 2003). The semantc trgger features consst of some specal head nouns for an entty class whch s suppled by users. Furthermore, a wndow (sze = 7), whch represents the local context of the target word w, s also used to classfy w. However, for actve learnng n NER, t s not reasonable to select a sngle word wthout context for human to label. Even f we requre human to label a sngle word, he has to make an addton effort to refer to the context of the word. In our actve learnng process, we select a word sequence whch conssts of a machne-annotated named entty and ts context rather than a sngle word. Therefore, all of the measures we propose for actve learnng should be appled to the machneannotated named enttes and we have to further study how to extend the measures for words to named enttes. Thus, the actve learnng n SVMbased NER wll be more complex than that n smple classfcaton tasks, such as text classf caton on whch most SVM actve learnng works are conducted (Schohn and Cohn 2000; Tong and Koller 2000; Brnker 2003). In the next part, we wll ntroduce nformatveness, representatveness and dversty measures for the SVM-based NER. 2.1 Informatveness The basc dea of nformatveness crteron s smlar to certanty-based sample selecton methods, whch have been used n many prevous works. In our task, we use a dstance-based measure to evaluate the nformatveness of a word and extend t to the measure of an entty usng three scorng functons. We prefer the examples wth hgh nformatve degree for whch the current model are most uncertan Informatveness Measure for Word In the smplest lnear form, tranng SVM s to fnd a hyperplane that can separate the post ve and negatve examples n tranng set wth maxmum margn. The margn s defned by the dstance of the hyperplane to the nearest of the postve and negatve examples. The tranng examples whch are closest to the hyperplane are called support vectors. In SVM, only the support vectors are useful for the classfcaton, whch s dfferent from statstcal models. SVM tranng s to get these support vectors and ther weghts from tranng set by solvng quadratc programmng problem. The support vectors can later be used to classfy the test data. Intutvely, we consder the nformatveness of an example as how t can make effect on the support vectors by addng t to tranng set. An example may be nformatve for the learner f the dstance of ts feature vector to the hyperplane s less than that of the support vectors to the hyperplane (equal to 1). Ths ntuton s also ustfed by (Schohn and Cohn 2000; Tong and Koller 2000) based on a verson space analyss. They state that labelng an example that les on or close to the hyperplane s guaranteed to have an effect on the soluton. In our task, we use the dstance to measure the nformatveness of an example. The dstance of a word s feature vector to the hyperplane s computed as follows: N Dst( w) = α yk( s, w ) + b = 1 where w s the feature vector of the word, a, y, s corresponds to the weght, the class and the feature vector of the th support vector respectvely. N s the number of the support vectors n current model. We select the example wth mnmal Dst, whch ndcates that t comes closest to the hyperplane n feature space. Ths example s consdered most nformatve for current model Informatveness Measure for Named Entty

3 Based on the above nformatveness measure for a word, we compute the overall nformatveness degree of a named entty NE. In ths paper, we propose three scorng functons as follows. Let NE = w 1 w N n whch w s the feature vector of the th word of NE. Info_Avg: The nformatveness of NE s scored by the average dstance of the words n NE to the hyperplane. Info( NE) = 1 Dst( w ) w NE where, w s the feature vector of the th word n NE. Info_Mn: The nformatveness of NE s scored by the mnmal dstance of the words n NE. Info( NE) = 1 Mn{ Dst( w )} w NE Info_S/N: If the dstance of a word to the hyperplane s less than a threshold a (= 1 n our task), the word s consdered wth short dstance. Then, we compute the proporton of the number of words wth short dstance to the total number of words n the named entty and use ths proporton to quantfy the nformatveness of the named entty. NUM ( Dst( w ) < α ) w NE Info( NE) = N In Secton 4.3, we wll evaluate the effectveness of these scorng functons. 2.2 Representatveness In addton to the most nformatve example, we also prefer the most representatve example. The representatveness of an example can be evaluated based on how many examples there are smlar or near to t. So, the examples wth hgh representatve degree are less lkely to be an outler. Addng them to the tranng set wll have effect on a large number of unlabeled examples. There are only a few works consderng ths selecton crteron (McCallum and Ngam 1998; Tang et al. 2002) and both of them are specfc to ther tasks, vz. text classfcaton and statstcal parsng. In ths secton, we compute the smlarty between words usng a general vector-based measure, extend ths measure to named entty level usng dynamc tme warpng algorthm and quantfy the representatveness of a named entty by ts densty Smlarty Measure between Words In general vector space model, the smlarty between two vectors may be measured by computng the cosne value of the angle between them. The smaller the angle s, the more smlar between the vectors are. Ths measure, called cosne-smlarty measure, has been wdely used n nformaton retreval tasks (Baeza-Yates and Rbero-Neto 1999). In our task, we also use t to quantfy the smlarty between two words. Partcularly, the calculaton n SVM need be proected to a hgher dmensonal space by usng a certan kernel functon K ( w, w ). Therefore, we adapt the cosne-smlarty measure to SVM as follows: Sm ( w, w ) = k ( w, w ) k( w, w ) k( w, w ) where, w and w are the feature vectors of the words and. Ths calculaton s also supported by (Brnker 2003) s work. Furthermore, f we use the lnear kernel k ( w, w ) = w w, the measure s the same as the tradtonal cosne smlarty measure cosθ = and may be regarded as a w w w w general vector-based smlarty measure Smlarty Meas ure between Named Enttes In ths part, we compute the smlarty between two machne-annotated named enttes gven the smlartes between words. Regardng an entty as a word sequence, ths work s analogous to the algnment of two sequences. We employ the dynamc tme warpng (DTW) algorthm (Rabner et al. 1978) to fnd an optmal algnment between the words n the sequences whch maxmze the accumulated smlarty degree between the sequences. Here, we adapt t to our task. A sketch of the modfed algorthm s as follows. Let NE 1 = w 11 w 12 w 1n w 1N, (n = 1,, N) and NE 2 = w 21 w 22 w 2m w 2M, (m = 1,, M) denote two word sequences to be matched. NE 1 and NE 2 consst of M and N words respectvely. NE 1 (n) = w 1n and NE 2 (m) = w 2m. A smlarty value Sm(w 1n,w 2m ) has been known for every par of words (w 1n,w 2m ) wthn NE 1 and NE 2. The goal of DTW s to fnd a path, m = map(n), whch map n onto the correspondng m such that the accumulated smlarty Sm* along the path s maxmzed. N Sm* = Max { Sm( NE1( n), NE 2( map( n))} { map( n)} n = 1 A dynamc programmng method s used to determne the optmum path map(n). The accumulated smlarty Sm A to any grd pont (n, m) can be recursvely calculated as Sm ( n, m) = Sm( w, w ) + MaxSm ( n 1, q) A 1n 2m A q m Fnally, Sm* = Sm A ( N, M ) Certanly, the overall smlarty measure Sm* has to be normalzed as longer sequences normally gve hgher smlarty value. So, the smlarty between two sequences NE 1 and NE 2 s calculated as

4 Sm * Sm( NE1, NE2) = Max( N, M ) Representatveness Measure for Named Entty Gven a set of machne-annotated named enttes NESet = {NE 1,, NE N }, the representatveness of a named entty NE n NESet s quantfed by ts densty. The densty of NE s defned as the average smlarty between NE and all the other enttes NE n NESet as follows. Sm( NE, NE ) Densty( NE) = N 1 If NE has the largest densty among all the enttes n NESet, t can be regarded as the centrod of NE- Set and also the most representatve examples n NESet. 2.3 Dversty Dversty crteron s to maxmze the tranng utlty of a batch. We prefer the batch n whch the examples have hgh varance to each other. For example, gven the batch sze 5, we try not to select fve repettous examples at a tme. To our knowledge, there s only one work (Brnker 2003) explorng ths crteron. In our task, we propose two methods: local and global, to make the examples dverse enough n a batch Global Consderaton For a global consderaton, we cluster all named enttes n NESet based on the smlarty measure proposed n Secton The named enttes n the same cluster may be consdered smlar to each other, so we wll select the named enttes from dfferent clusters at one tme. We employ a K- means clusterng algorthm (Jelnek 1997), whch s shown n Fgure 1. Gven: NESet = {NE 1,, NE N } Suppose: The number of clusters s K Intalzaton: Randomly equally partton {NE 1,, NE N } nto K ntal clusters C ( = 1,, K). Loop untl the number of changes for the centrods of all clusters s less than a threshold Fnd the centrod of each cluster C ( = 1,, K). NECent = arg max( Sm( NE, NE)) NE C NE C Repartton {NE 1,, NE N } nto K clusters. NE wll be assgned to Cluster C f Sm( NE, NECent ) Sm( NE, NECent ), w w Fgure 1: Global Consderaton for Dversty: K- Means Clusterng algorthm In each round, we need to compute the parwse smlartes wthn each cluster to get the centrod of the cluster. And then, we need to compute the smlartes between each example and all centrods to repartton the examples. So, the algorthm s tme-consumng. Based on the assumpton that N examples are unformly dstrbuted between the K clusters, the tme complexty of the algorthm s about O(N 2 /K+NK) (Tang et al. 2002). In one of our experments, the sze of the NESet (N) s around and K s equal to 50, so the tme complexty s about O(10 6 ). For effcency, we may flter the enttes n NESet before clusterng them, whch wll be further dscussed n Secton Local Consderaton When selectng a machne-annotated named entty, we compare t wth all prevously selected named enttes n the current batch. If the smlarty between them s above a threshold ß, ths example cannot be allowed to add nto the batch. The order of selectng examples s based on some measure, such as nformatveness measure, representatveness measure or ther combnaton. Ths local selecton method s shown n Fgure 2. In ths way, we avod selectng too smlar examples (smlarty value ß) n a batch. The threshold ß may be the average smlarty between the examples n NESet. Gven: NESet = {NE 1,, NE N } BatchSet wth the maxmal sze K. Intalzaton: BatchSet = empty Loop untl BatchSet s full Select NE based on some measure from NESet. RepeatFlag = false; Loop from = 1 to CurrentSze(BatchSet) If Sm( NE, NE ) β Then RepeatFlag = true; Stop the Loop; If RepeatFlag == false Then add NE nto BatchSet remove NE from NESet Fgure 2: Local Consderaton for Dversty Ths consderaton only requres O(NK+K 2 ) computatonal tme. In one of our experments (N and K = 50), the tme complexty s about O(10 5 ). It s more effcent than clusterng alg o- rthm descrbed n Secton Sample Selecton strateges In ths secton, we wll study how to combne and strke a proper balance between these crtera, vz. nformatveness, representatveness and dversty,

5 to reach the maxmum effectveness on NER actve learnng. We buld two strateges to combne the measures proposed above. These strateges are based on the varyng prortes of the crtera and the varyng degrees to satsfy the crtera. Strategy 1: We frst consder the nformatveness crteron. We choose m examples wth the most nformatveness score from NESet to an ntermedate set called INTERSet. By ths preselectng, we make the selecton process faster n the later steps snce the sze of INTERSet s much smaller than that of NESet. Then we cluster the examples n INTERSet and choose the centrod of each cluster nto a batch called BatchSet. The centrod of a cluster s the most representatve example n that cluster snce t has the largest densty. Furthermore, the examples n dfferent clusters may be consdered dverse to each other. By ths means, we consder representatveness and dversty crtera at the same tme. Ths strategy s shown n Fgure 3. One lmtaton of ths strategy s that clusterng result may not reflect the dstrb u- ton of whole sample space snce we only cluster on INTERSet for effcency. The other s that snce the representatveness of an example s only evaluated on a cluster. If the cluster sze s too small, the most representatve example n ths cluster may not be representatve n the whole sample space. NE wth the maxmum value of ths functon from NESet. Second, we consder dversty crteron usng the local method n Secton We add the canddate example NE to a batch only f NE s dfferent enough from any prevously selected example n the batch. The threshold ß s set to the average par-wse smlarty of the enttes n NE- Set. Gven: NESet = {NE 1,, NE N } BatchSet wth the maxmal sze K. Intalzaton: BatchSet = Loop untl BatchSet s full Select NE whch have the maxmum value for the combnaton functon between Info score and Densty socre from NESet. NE = arg Max( λinfo( NE ) + (1 λ) Densty( NE )) NE NESet RepeatFlag = false; Loop from = 1 to CurrentSze(BatchSet) If Sm( NE, NE ) β Then RepeatFlag = true; Stop the Loop; If RepeatFlag == false Then add NE nto BatchSet remove NE from NESet Fgure 4: Sample Selecton Strategy 2 Gven: NESet = {NE 1,, NE N } BatchSet wth the maxmal sze K. INTERSet wth the maxmal sze M Steps : BatchSet = INTERSet = Select M enttes wth most Info score from NESet to INTERSet. Cluster the enttes n INTERSet nto K clusters Add the centrod entty of each cluster to BatchSet Fgure 3: Sample Selecton Strategy 1 Strategy 2: (Fgure 4) We combne the nformatveness and representatveness crtera usng the functo λinfo( NE) + (1 λ) Densty( NE), n whch the Info and Densty value of NE are normalzed frst. The ndvdual mportance of each crteron n ths functon s adusted by the tradeoff parameter λ ( 0 λ 1 ) (set to 0.6 n our experment). Frst, we select a canddate example 4 Expermental Results and Analyss 4.1 Experment Settngs In order to evaluate the effectveness of our sele c- ton strateges, we apply them to recognze proten (PRT) names n bomedcal doman usng GENIA corpus V1.1 (Ohta et al. 2002) and person (PER), locaton (LOC), organzaton (ORG) names n newswre doman usng MUC-6 corpus. Frst, we randomly splt the whole corpus nto three parts: an ntal tranng set to buld an ntal model, a test set to evaluate the performance of the model and an unlabeled set to select examples. The sze of each data set s shown n Table 1. Then, teratvely, we select a batch of examples followng the selecton strateges proposed, requre human experts to label them and add them nto the tranng set. The batch sze K = 50 n GENIA and 10 n MUC-6. Each example s defned as a machne-recognzed named entty and ts context words (prevous 3 words and next 3 words). Doman Class Corpus Intal Tranng Set Test Set Unlabeled Set Bomedcal PRT GENIA sent. (277 words) 900 sent. (26K words) 8004 sent. (223K words) PER 5 sent. (131 words) 7809 sent. (157K words) Newswre LOC MUC-6 5 sent. (130 words) 602 sent. (14K words) 7809 sent. (157K words) ORG 5 sent. (113 words) 7809 sent. (157K words) Table 1: Experment settngs for actve learnng usng GENIA1.1(PRT) and MUC-6(PER,LOC,ORG)

6 The goal of our work s to mnmze the human annotaton effort to learn a named entty recognzer wth the same performance level as supervsed learnng. The performance of our model s evaluated usng precson/recall/f-measure. 4.2 Overall Result n GENIA and MUC-6 In ths secton, we evaluate our selecton strateges by comparng them wth a random selecton method, n whch a batch of examples s randomly selected teratvely, on GENIA and MUC-6 corpus. Table 2 shows the amount of tranng data needed to acheve the performance of supervsed learnng usng varous selecton methods, vz. Random, Strategy1 and Strategy2. In GENIA, we fnd: The model acheves 63.3 F-measure usng 223K words n the supervsed learnng. The best performer s Strategy2 (31K words), requrng less than 40% of the tranng data that Random (83K words) does and 14% of the tranng data that the supervsed learnng does. Strategy1 (40K words) performs slghtly worse than Strategy2, requrng 9K more words. It s probably because Strategy1 cannot avod selectng outlers f a cluster s too small. Random (83K words) requres about 37% of the tranng data that the supervsed learnng does. It ndcates that only the words n and around a named entty are useful for classfcaton and the words far from the named entty may not be helpful. Class Supervsed Random Strategy1 Strategy2 PRT 223K (F=63.3) 83K 40K 31K PER 157K (F=90.4) 11.5K 4.2K 3.5K LOC 157K (F=73.5) 13.6K 3.5K 2.1K ORG 157K (F=86.0) 20.2K 9.5K 7.8K Table 2: Overall Result n GENIA and MUC-6 Furthermore, when we apply our model to newswre doman (MUC-6) to recognze person, locaton and organzaton names, Strategy1 and Strategy2 show a more promsng result by comparng wth the supervsed learnng and Random, as shown n Table 2. On average, about 95% of the data can be reduced to acheve the same performance wth the supervsed learnng n MUC-6. It s probably because NER n the newswre doman s much smpler than that n the bomedcal doman (Shen et al. 2003) and named enttes are less and dstrbuted much sparser n the newswre texts than n the bomedcal texts. 4.3 Effectveness of Informatveness-based Selecton Method In ths secton, we nvestgate the effectveness of nformatveness crteron n NER task. Fgure 5 shows a plot of tranng data sze versus F-measure acheved by the nformatveness-based measures n Secton 3.1.2: Info_Avg, Info_Mn and Info_S/N as well as Random. We make the comparsons n GENIA corpus. In Fgure 5, the horzontal lne s the performance level (63.3 F-measure) acheved by supervsed learnng (223K words). We fnd that the three nformatveness-based measures perform smlarly and each of them outperforms Random. Table 3 hghlghts the varous data szes to acheve the peak performance usng these selecton methods. We fnd that Random (83K words) on average requres over 1.5 tmes as much as data to acheve the same performance as the nformatveness-based selecton methods (52K words) F Supervsed Random Info_Mn Info_S/N Info_Avg K words 80 Fgure 5: Actve learnng curves: effectveness of the three nformatveness-crteron-based selectons comparng wth the Random selecton. Supervsed Random Info_Avg Info_Mn Info_ S/N 223K 83K 52.0K 51.9K 52.3K Table 3: Tranng data szes for varous selecton methods to acheve the same performance level as the supervsed learnng 4.4 Effectveness of Two Sample Selecton Strateges In addton to the nformatveness crteron, we further ncorporate representatveness and dversty crtera nto actve learnng usng two strateges descrbed n Secton 3. Comparng the two strateges wth the best result of the sngle-crteronbased selecton methods Info_Mn, we are to ustfy that representatveness and dversty are also mportant factors for actve learnng. Fgure 6 shows the learnng curves for the varous methods: Strategy1, Strategy2 and Info_Mn. In the begnnng teratons (F-measure < 60), the three methods performed smlarly. But wth the larger tranng set, the effcences of Stratety1 and Strategy2 begn to be evdent. Table 4 hghlghts the fnal result of the three methods. In order to reach the performance of supervsed learnng, Strategy1 (40K words) and Strategyy2 (31K words) requre about 80% and 60% of the data that Info_Mn (51.9K) does. So we beleve the effectve combnatons of nformatveness, representatveness and dversty wll help to learn the NER model more quckly and cost less n annotaton.

7 0.65 F Supervsed Info_Mn Strategy1 Strategy K words Fgure 6: Actve learnng curves: effectveness of the two mult-crtera-based selecton strateges comparng wth the nformatveness-crteron-based selecton (Info_Mn). Info_Mn Strategy1 Strategy2 51.9K 40K 31K Table 4: Comparsons of tranng data szes for the multcrtera-based selecton strateges and the nformatvenesscrteron-based selecton (Info_Mn) to acheve the same performance level as the supervsed learnng. 5 Related Work Snce there s no study on actve learnng for NER task prevously, we only ntroduce general actve learnng methods here. Many exstng actve learnng methods are to select the most uncertan examples usng varous measures (Thompson et al. 1999; Schohn and Cohn 2000; Tong and Koller 2000; Engelson and Dagan 1999; Nga and Yarowsky 2000). Our nformatveness-based measure s smlar to these works. However these works ust follow a sngle crteron. (McCallum and Ngam 1998; Tang et al. 2002) are the only two works consderng the representatveness crteron n actve learnng. (Tang et al. 2002) use the densty nformaton to weght the selected examples whle we use t to select examples. Moreover, the representatveness measure we use s relatvely general and easy to adapt to other tasks, n whch the example selected s a sequence of words, such as text chunkng, POS taggng, etc. On the other hand, (Brnker 2003) frst ncorporate dversty n actve learnng for text classfcaton. Ther work s smlar to our local consderaton n Secton However, he ddn t further explore how to avod selectng outlers to a batch. So far, we haven t found any prevous work ntegratng the nformatveness, representatveness and dversty all together. 6 Concluson and Future Work In ths paper, we study the actve learnng n a more complex NLP task, named entty recognton. We propose a mult-crtera-based approach to select examples based on ther nformatveness, representatveness and dversty, whch are ncorporated all together by two strateges (local and global). Experments show that, n both MUC- 6 and GENIA, both of the two strateges combnng the three crtera outperform the sngle crteron (nformatveness). The labelng cost can be sgnfcantly reduced by at least 80% comparng wth the supervsed learnng. To our best knowledge, ths s not only the frst work to report the emprcal results of actve learnng for NER, but also the frst work to ncorporate the three crtera all together for selectng examples. Although the current experment results are very promsng, some parameters n our experment, such as the batch sze K and the λ n the functon of strategy 2, are decded by our experence n the doman. In practcal applcaton, the optmal value of these parameters should be decded automatcally based on the tranng process. Furthermore, we wll study how to overcome the lmtaton of the strategy 1 dscussed n Secton 3 by usng more effectve clusterng algorthm. Another nterestng work s to study when to stop actve learnng. References R. Baeza-Yates and B. Rbero-Neto Modern Informaton Retreval. ISBN X. K. Brnker Incorporatng Dversty n Actve Learnng wth Support Vector Machnes. In Proceedngs of ICML, S. A. Engelson and I. Dagan Commttee- Based Sample Selecton for Probablstc Classfers. Journal of Artfcal Intellgence Research. F. Jelnek Statstcal Methods for Speech Recognton. MIT Press. J. Kazama, T. Makno, Y. Ohta and J. Tsu Tunng Support Vector Machnes for Bomedcal Named Entty Recognton. In Proceedngs of the ACL2002 Workshop on NLP n Bomedcne. K. J. Lee, Y. S. Hwang and H. C. Rm Two- Phase Bomedcal NE Recognton based on SVMs. In Proceedngs of the ACL2003 Workshop on NLP n Bomedcne. D. D. Lews and J. Catlett Heterogeneous Uncertanty Samplng for Supervsed Learnng. In Proceedngs of ICML, A. McCallum and K. Ngam Employng EM n Pool-Based Actve Learnng for Text Classfcaton. In Proceedngs of ICML, G. Nga and D. Yarowsky Rule Wrtng or Annotaton: Cost-effcent Resource Usage for Base Noun Phrase Chunkng. In Proceedngs of ACL, 2000.

8 T. Ohta, Y. Tates, J. Km, H. Mma and J. Tsu The GENIA corpus: An annotated research abstract corpus n molecular bology doman. In Proceedngs of HLT L. R. Rabner, A. E. Rosenberg and S. E. Levnson Consderatons n Dynamc Tme Warpng Algorthms for Dscrete Word Recognton. In Proceedngs of IEEE Transactons on acoustcs, speech and sgnal processng. Vol. ASSP-26, NO.6. D. Schohn and D. Cohn Less s More: Actve Learnng wth Support Vector Machnes. In Proceedngs of the 17 th Internatonal Conference on Machne Learnng. D. Shen, J. Zhang, G. D. Zhou, J. Su and C. L. Tan Effectve Adaptaton of a Hdden Markov Model-based Named Entty Recognzer for Bomedcal Doman. In Proceedngs of the ACL2003 Workshop on NLP n Bomedcne. M. Steedman, R. Hwa, S. Clark, M. Osborne, A. Sarkar, J. Hockenmaer, P. Ruhlen, S. Baker and J. Crm Example Selecton for Bootstrappng Statstcal Parsers. In Proceedngs of HLT- NAACL, M. Tang, X. Luo and S. Roukos Actve Learnng for Statstcal Natural Language Parsng. In Proceedngs of the ACL C. A. Thompson, M. E. Calff and R. J. Mooney Actve Learnng for Natural Language Parsng and Informaton Extracton. In Proceedngs of ICML S. Tong and D. Koller Support Vector Machne Actve Learnng wth Applcatons to Text Classfcaton. Journal of Machne Learnng Research. V. Vapnk Statstcal learnng theory. N.Y.:John Wley.

Support Vector Machines

Support Vector Machines /9/207 MIST.6060 Busness Intellgence and Data Mnng What are Support Vector Machnes? Support Vector Machnes Support Vector Machnes (SVMs) are supervsed learnng technques that analyze data and recognze patterns.

More information

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1 4/14/011 Outlne Dscrmnatve classfers for mage recognton Wednesday, Aprl 13 Krsten Grauman UT-Austn Last tme: wndow-based generc obect detecton basc ppelne face detecton wth boostng as case study Today:

More information

The Research of Support Vector Machine in Agricultural Data Classification

The Research of Support Vector Machine in Agricultural Data Classification The Research of Support Vector Machne n Agrcultural Data Classfcaton Le Sh, Qguo Duan, Xnmng Ma, Me Weng College of Informaton and Management Scence, HeNan Agrcultural Unversty, Zhengzhou 45000 Chna Zhengzhou

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Decson surface s a hyperplane (lne n 2D) n feature space (smlar to the Perceptron) Arguably, the most mportant recent dscovery n machne learnng In a nutshell: map the data to a predetermned

More information

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Parallelism for Nested Loops with Non-uniform and Flow Dependences Parallelsm for Nested Loops wth Non-unform and Flow Dependences Sam-Jn Jeong Dept. of Informaton & Communcaton Engneerng, Cheonan Unversty, 5, Anseo-dong, Cheonan, Chungnam, 330-80, Korea. seong@cheonan.ac.kr

More information

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance

Tsinghua University at TAC 2009: Summarizing Multi-documents by Information Distance Tsnghua Unversty at TAC 2009: Summarzng Mult-documents by Informaton Dstance Chong Long, Mnle Huang, Xaoyan Zhu State Key Laboratory of Intellgent Technology and Systems, Tsnghua Natonal Laboratory for

More information

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers IOSR Journal of Electroncs and Communcaton Engneerng (IOSR-JECE) e-issn: 78-834,p- ISSN: 78-8735.Volume 9, Issue, Ver. IV (Mar - Apr. 04), PP 0-07 Content Based Image Retreval Usng -D Dscrete Wavelet wth

More information

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points; Subspace clusterng Clusterng Fundamental to all clusterng technques s the choce of dstance measure between data ponts; D q ( ) ( ) 2 x x = x x, j k = 1 k jk Squared Eucldean dstance Assumpton: All features

More information

Classifier Selection Based on Data Complexity Measures *

Classifier Selection Based on Data Complexity Measures * Classfer Selecton Based on Data Complexty Measures * Edth Hernández-Reyes, J.A. Carrasco-Ochoa, and J.Fco. Martínez-Trndad Natonal Insttute for Astrophyscs, Optcs and Electroncs, Lus Enrque Erro No.1 Sta.

More information

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Learning the Kernel Parameters in Kernel Minimum Distance Classifier Learnng the Kernel Parameters n Kernel Mnmum Dstance Classfer Daoqang Zhang 1,, Songcan Chen and Zh-Hua Zhou 1* 1 Natonal Laboratory for Novel Software Technology Nanjng Unversty, Nanjng 193, Chna Department

More information

CAN COMPUTERS LEARN FASTER? Seyda Ertekin Computer Science & Engineering The Pennsylvania State University

CAN COMPUTERS LEARN FASTER? Seyda Ertekin Computer Science & Engineering The Pennsylvania State University CAN COMPUTERS LEARN FASTER? Seyda Ertekn Computer Scence & Engneerng The Pennsylvana State Unversty sertekn@cse.psu.edu ABSTRACT Ever snce computers were nvented, manknd wondered whether they mght be made

More information

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification Introducton to Artfcal Intellgence V22.0472-001 Fall 2009 Lecture 24: Nearest-Neghbors & Support Vector Machnes Rob Fergus Dept of Computer Scence, Courant Insttute, NYU Sldes from Danel Yeung, John DeNero

More information

Machine Learning: Algorithms and Applications

Machine Learning: Algorithms and Applications 14/05/1 Machne Learnng: Algorthms and Applcatons Florano Zn Free Unversty of Bozen-Bolzano Faculty of Computer Scence Academc Year 011-01 Lecture 10: 14 May 01 Unsupervsed Learnng cont Sldes courtesy of

More information

A Deflected Grid-based Algorithm for Clustering Analysis

A Deflected Grid-based Algorithm for Clustering Analysis A Deflected Grd-based Algorthm for Clusterng Analyss NANCY P. LIN, CHUNG-I CHANG, HAO-EN CHUEH, HUNG-JEN CHEN, WEI-HUA HAO Department of Computer Scence and Informaton Engneerng Tamkang Unversty 5 Yng-chuan

More information

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning Outlne Artfcal Intellgence and ts applcatons Lecture 8 Unsupervsed Learnng Professor Danel Yeung danyeung@eee.org Dr. Patrck Chan patrckchan@eee.org South Chna Unversty of Technology, Chna Introducton

More information

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data A Fast Content-Based Multmeda Retreval Technque Usng Compressed Data Borko Furht and Pornvt Saksobhavvat NSF Multmeda Laboratory Florda Atlantc Unversty, Boca Raton, Florda 3343 ABSTRACT In ths paper,

More information

Determining the Optimal Bandwidth Based on Multi-criterion Fusion

Determining the Optimal Bandwidth Based on Multi-criterion Fusion Proceedngs of 01 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 5 (01) (01) IACSIT Press, Sngapore Determnng the Optmal Bandwdth Based on Mult-crteron Fuson Ha-L Lang 1+, Xan-Mn

More information

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization Problem efntons and Evaluaton Crtera for Computatonal Expensve Optmzaton B. Lu 1, Q. Chen and Q. Zhang 3, J. J. Lang 4, P. N. Suganthan, B. Y. Qu 6 1 epartment of Computng, Glyndwr Unversty, UK Faclty

More information

Syntactic Tree-based Relation Extraction Using a Generalization of Collins and Duffy Convolution Tree Kernel

Syntactic Tree-based Relation Extraction Using a Generalization of Collins and Duffy Convolution Tree Kernel Syntactc Tree-based Relaton Extracton Usng a Generalzaton of Collns and Duffy Convoluton Tree Kernel Mahdy Khayyaman Seyed Abolghasem Hassan Abolhassan Mrroshandel Sharf Unversty of Technology Sharf Unversty

More information

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law)

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law) Machne Learnng Support Vector Machnes (contans materal adapted from talks by Constantn F. Alfers & Ioanns Tsamardnos, and Martn Law) Bryan Pardo, Machne Learnng: EECS 349 Fall 2014 Support Vector Machnes

More information

Efficient Text Classification by Weighted Proximal SVM *

Efficient Text Classification by Weighted Proximal SVM * Effcent ext Classfcaton by Weghted Proxmal SVM * Dong Zhuang 1, Benyu Zhang, Qang Yang 3, Jun Yan 4, Zheng Chen, Yng Chen 1 1 Computer Scence and Engneerng, Bejng Insttute of echnology, Bejng 100081, Chna

More information

Incremental Learning with Support Vector Machines and Fuzzy Set Theory

Incremental Learning with Support Vector Machines and Fuzzy Set Theory The 25th Workshop on Combnatoral Mathematcs and Computaton Theory Incremental Learnng wth Support Vector Machnes and Fuzzy Set Theory Yu-Mng Chuang 1 and Cha-Hwa Ln 2* 1 Department of Computer Scence and

More information

TN348: Openlab Module - Colocalization

TN348: Openlab Module - Colocalization TN348: Openlab Module - Colocalzaton Topc The Colocalzaton module provdes the faclty to vsualze and quantfy colocalzaton between pars of mages. The Colocalzaton wndow contans a prevew of the two mages

More information

A Unified Framework for Semantics and Feature Based Relevance Feedback in Image Retrieval Systems

A Unified Framework for Semantics and Feature Based Relevance Feedback in Image Retrieval Systems A Unfed Framework for Semantcs and Feature Based Relevance Feedback n Image Retreval Systems Ye Lu *, Chunhu Hu 2, Xngquan Zhu 3*, HongJang Zhang 2, Qang Yang * School of Computng Scence Smon Fraser Unversty

More information

Object-Based Techniques for Image Retrieval

Object-Based Techniques for Image Retrieval 54 Zhang, Gao, & Luo Chapter VII Object-Based Technques for Image Retreval Y. J. Zhang, Tsnghua Unversty, Chna Y. Y. Gao, Tsnghua Unversty, Chna Y. Luo, Tsnghua Unversty, Chna ABSTRACT To overcome the

More information

CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION

CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION 48 CHAPTER 3 SEQUENTIAL MINIMAL OPTIMIZATION TRAINED SUPPORT VECTOR CLASSIFIER FOR CANCER PREDICTION 3.1 INTRODUCTION The raw mcroarray data s bascally an mage wth dfferent colors ndcatng hybrdzaton (Xue

More information

Manifold-Ranking Based Keyword Propagation for Image Retrieval *

Manifold-Ranking Based Keyword Propagation for Image Retrieval * Manfold-Rankng Based Keyword Propagaton for Image Retreval * Hanghang Tong,, Jngru He,, Mngjng L 2, We-Yng Ma 2, Hong-Jang Zhang 2 and Changshu Zhang 3,3 Department of Automaton, Tsnghua Unversty, Bejng

More information

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms Course Introducton Course Topcs Exams, abs, Proects A quc loo at a few algorthms 1 Advanced Data Structures and Algorthms Descrpton: We are gong to dscuss algorthm complexty analyss, algorthm desgn technques

More information

Outline. Self-Organizing Maps (SOM) US Hebbian Learning, Cntd. The learning rule is Hebbian like:

Outline. Self-Organizing Maps (SOM) US Hebbian Learning, Cntd. The learning rule is Hebbian like: Self-Organzng Maps (SOM) Turgay İBRİKÇİ, PhD. Outlne Introducton Structures of SOM SOM Archtecture Neghborhoods SOM Algorthm Examples Summary 1 2 Unsupervsed Hebban Learnng US Hebban Learnng, Cntd 3 A

More information

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

Classifying Acoustic Transient Signals Using Artificial Intelligence

Classifying Acoustic Transient Signals Using Artificial Intelligence Classfyng Acoustc Transent Sgnals Usng Artfcal Intellgence Steve Sutton, Unversty of North Carolna At Wlmngton (suttons@charter.net) Greg Huff, Unversty of North Carolna At Wlmngton (jgh7476@uncwl.edu)

More information

Optimizing Document Scoring for Query Retrieval

Optimizing Document Scoring for Query Retrieval Optmzng Document Scorng for Query Retreval Brent Ellwen baellwe@cs.stanford.edu Abstract The goal of ths project was to automate the process of tunng a document query engne. Specfcally, I used machne learnng

More information

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching A Fast Vsual Trackng Algorthm Based on Crcle Pxels Matchng Zhqang Hou hou_zhq@sohu.com Chongzhao Han czhan@mal.xjtu.edu.cn Ln Zheng Abstract: A fast vsual trackng algorthm based on crcle pxels matchng

More information

Intelligent Information Acquisition for Improved Clustering

Intelligent Information Acquisition for Improved Clustering Intellgent Informaton Acquston for Improved Clusterng Duy Vu Unversty of Texas at Austn duyvu@cs.utexas.edu Mkhal Blenko Mcrosoft Research mblenko@mcrosoft.com Prem Melvlle IBM T.J. Watson Research Center

More information

Query Clustering Using a Hybrid Query Similarity Measure

Query Clustering Using a Hybrid Query Similarity Measure Query clusterng usng a hybrd query smlarty measure Fu. L., Goh, D.H., & Foo, S. (2004). WSEAS Transacton on Computers, 3(3), 700-705. Query Clusterng Usng a Hybrd Query Smlarty Measure Ln Fu, Don Hoe-Lan

More information

CS 534: Computer Vision Model Fitting

CS 534: Computer Vision Model Fitting CS 534: Computer Vson Model Fttng Sprng 004 Ahmed Elgammal Dept of Computer Scence CS 534 Model Fttng - 1 Outlnes Model fttng s mportant Least-squares fttng Maxmum lkelhood estmaton MAP estmaton Robust

More information

Announcements. Supervised Learning

Announcements. Supervised Learning Announcements See Chapter 5 of Duda, Hart, and Stork. Tutoral by Burge lnked to on web page. Supervsed Learnng Classfcaton wth labeled eamples. Images vectors n hgh-d space. Supervsed Learnng Labeled eamples

More information

UB at GeoCLEF Department of Geography Abstract

UB at GeoCLEF Department of Geography   Abstract UB at GeoCLEF 2006 Mguel E. Ruz (1), Stuart Shapro (2), June Abbas (1), Slva B. Southwck (1) and Davd Mark (3) State Unversty of New York at Buffalo (1) Department of Lbrary and Informaton Studes (2) Department

More information

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique //00 :0 AM Outlne and Readng The Greedy Method The Greedy Method Technque (secton.) Fractonal Knapsack Problem (secton..) Task Schedulng (secton..) Mnmum Spannng Trees (secton.) Change Money Problem Greedy

More information

A PATTERN RECOGNITION APPROACH TO IMAGE SEGMENTATION

A PATTERN RECOGNITION APPROACH TO IMAGE SEGMENTATION 1 THE PUBLISHING HOUSE PROCEEDINGS OF THE ROMANIAN ACADEMY, Seres A, OF THE ROMANIAN ACADEMY Volume 4, Number 2/2003, pp.000-000 A PATTERN RECOGNITION APPROACH TO IMAGE SEGMENTATION Tudor BARBU Insttute

More information

Keyword-based Document Clustering

Keyword-based Document Clustering Keyword-based ocument lusterng Seung-Shk Kang School of omputer Scence Kookmn Unversty & AIrc hungnung-dong Songbuk-gu Seoul 36-72 Korea sskang@kookmn.ac.kr Abstract ocument clusterng s an aggregaton of

More information

Edge Detection in Noisy Images Using the Support Vector Machines

Edge Detection in Noisy Images Using the Support Vector Machines Edge Detecton n Nosy Images Usng the Support Vector Machnes Hlaro Gómez-Moreno, Saturnno Maldonado-Bascón, Francsco López-Ferreras Sgnal Theory and Communcatons Department. Unversty of Alcalá Crta. Madrd-Barcelona

More information

GSLM Operations Research II Fall 13/14

GSLM Operations Research II Fall 13/14 GSLM 58 Operatons Research II Fall /4 6. Separable Programmng Consder a general NLP mn f(x) s.t. g j (x) b j j =. m. Defnton 6.. The NLP s a separable program f ts objectve functon and all constrants are

More information

Term Weighting Classification System Using the Chi-square Statistic for the Classification Subtask at NTCIR-6 Patent Retrieval Task

Term Weighting Classification System Using the Chi-square Statistic for the Classification Subtask at NTCIR-6 Patent Retrieval Task Proceedngs of NTCIR-6 Workshop Meetng, May 15-18, 2007, Tokyo, Japan Term Weghtng Classfcaton System Usng the Ch-square Statstc for the Classfcaton Subtask at NTCIR-6 Patent Retreval Task Kotaro Hashmoto

More information

Unsupervised Learning

Unsupervised Learning Pattern Recognton Lecture 8 Outlne Introducton Unsupervsed Learnng Parametrc VS Non-Parametrc Approach Mxture of Denstes Maxmum-Lkelhood Estmates Clusterng Prof. Danel Yeung School of Computer Scence and

More information

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET 1 BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET TZU-CHENG CHUANG School of Electrcal and Computer Engneerng, Purdue Unversty, West Lafayette, Indana 47907 SAUL B. GELFAND School

More information

Relevance Feedback Document Retrieval using Non-Relevant Documents

Relevance Feedback Document Retrieval using Non-Relevant Documents Relevance Feedback Document Retreval usng Non-Relevant Documents TAKASHI ONODA, HIROSHI MURATA and SEIJI YAMADA Ths paper reports a new document retreval method usng non-relevant documents. From a large

More information

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION

BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION BAYESIAN MULTI-SOURCE DOMAIN ADAPTATION SHI-LIANG SUN, HONG-LEI SHI Department of Computer Scence and Technology, East Chna Normal Unversty 500 Dongchuan Road, Shangha 200241, P. R. Chna E-MAIL: slsun@cs.ecnu.edu.cn,

More information

Collaboratively Regularized Nearest Points for Set Based Recognition

Collaboratively Regularized Nearest Points for Set Based Recognition Academc Center for Computng and Meda Studes, Kyoto Unversty Collaboratvely Regularzed Nearest Ponts for Set Based Recognton Yang Wu, Mchhko Mnoh, Masayuk Mukunok Kyoto Unversty 9/1/013 BMVC 013 @ Brstol,

More information

Machine Learning 9. week

Machine Learning 9. week Machne Learnng 9. week Mappng Concept Radal Bass Functons (RBF) RBF Networks 1 Mappng It s probably the best scenaro for the classfcaton of two dataset s to separate them lnearly. As you see n the below

More information

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices Internatonal Mathematcal Forum, Vol 7, 2012, no 52, 2549-2554 An Applcaton of the Dulmage-Mendelsohn Decomposton to Sparse Null Space Bases of Full Row Rank Matrces Mostafa Khorramzadeh Department of Mathematcal

More information

Load Balancing for Hex-Cell Interconnection Network

Load Balancing for Hex-Cell Interconnection Network Int. J. Communcatons, Network and System Scences,,, - Publshed Onlne Aprl n ScRes. http://www.scrp.org/journal/jcns http://dx.do.org/./jcns.. Load Balancng for Hex-Cell Interconnecton Network Saher Manaseer,

More information

Japanese Dependency Analysis Based on Improved SVM and KNN

Japanese Dependency Analysis Based on Improved SVM and KNN Proceedngs of the 7th WSEAS Internatonal Conference on Smulaton, Modellng and Optmzaton, Bejng, Chna, September 15-17, 2007 140 Japanese Dependency Analyss Based on Improved SVM and KNN ZHOU HUIWEI and

More information

Face Recognition Based on SVM and 2DPCA

Face Recognition Based on SVM and 2DPCA Vol. 4, o. 3, September, 2011 Face Recognton Based on SVM and 2DPCA Tha Hoang Le, Len Bu Faculty of Informaton Technology, HCMC Unversty of Scence Faculty of Informaton Scences and Engneerng, Unversty

More information

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION Paulo Quntlano 1 & Antono Santa-Rosa 1 Federal Polce Department, Brasla, Brazl. E-mals: quntlano.pqs@dpf.gov.br and

More information

Cluster Analysis of Electrical Behavior

Cluster Analysis of Electrical Behavior Journal of Computer and Communcatons, 205, 3, 88-93 Publshed Onlne May 205 n ScRes. http://www.scrp.org/ournal/cc http://dx.do.org/0.4236/cc.205.350 Cluster Analyss of Electrcal Behavor Ln Lu Ln Lu, School

More information

Face Recognition University at Buffalo CSE666 Lecture Slides Resources:

Face Recognition University at Buffalo CSE666 Lecture Slides Resources: Face Recognton Unversty at Buffalo CSE666 Lecture Sldes Resources: http://www.face-rec.org/algorthms/ Overvew of face recognton algorthms Correlaton - Pxel based correspondence between two face mages Structural

More information

3D vector computer graphics

3D vector computer graphics 3D vector computer graphcs Paolo Varagnolo: freelance engneer Padova Aprl 2016 Prvate Practce ----------------------------------- 1. Introducton Vector 3D model representaton n computer graphcs requres

More information

From Comparing Clusterings to Combining Clusterings

From Comparing Clusterings to Combining Clusterings Proceedngs of the Twenty-Thrd AAAI Conference on Artfcal Intellgence (008 From Comparng Clusterngs to Combnng Clusterngs Zhwu Lu and Yuxn Peng and Janguo Xao Insttute of Computer Scence and Technology,

More information

Pruning Training Corpus to Speedup Text Classification 1

Pruning Training Corpus to Speedup Text Classification 1 Prunng Tranng Corpus to Speedup Text Classfcaton Jhong Guan and Shugeng Zhou School of Computer Scence, Wuhan Unversty, Wuhan, 430079, Chna hguan@wtusm.edu.cn State Key Lab of Software Engneerng, Wuhan

More information

User Authentication Based On Behavioral Mouse Dynamics Biometrics

User Authentication Based On Behavioral Mouse Dynamics Biometrics User Authentcaton Based On Behavoral Mouse Dynamcs Bometrcs Chee-Hyung Yoon Danel Donghyun Km Department of Computer Scence Department of Computer Scence Stanford Unversty Stanford Unversty Stanford, CA

More information

Face Recognition Method Based on Within-class Clustering SVM

Face Recognition Method Based on Within-class Clustering SVM Face Recognton Method Based on Wthn-class Clusterng SVM Yan Wu, Xao Yao and Yng Xa Department of Computer Scence and Engneerng Tong Unversty Shangha, Chna Abstract - A face recognton method based on Wthn-class

More information

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 A mathematcal programmng approach to the analyss, desgn and

More information

The Study of Remote Sensing Image Classification Based on Support Vector Machine

The Study of Remote Sensing Image Classification Based on Support Vector Machine Sensors & Transducers 03 by IFSA http://www.sensorsportal.com The Study of Remote Sensng Image Classfcaton Based on Support Vector Machne, ZHANG Jan-Hua Key Research Insttute of Yellow Rver Cvlzaton and

More information

Experiments in Text Categorization Using Term Selection by Distance to Transition Point

Experiments in Text Categorization Using Term Selection by Distance to Transition Point Experments n Text Categorzaton Usng Term Selecton by Dstance to Transton Pont Edgar Moyotl-Hernández, Héctor Jménez-Salazar Facultad de Cencas de la Computacón, B. Unversdad Autónoma de Puebla, 14 Sur

More information

BRDPHHC: A Balance RDF Data Partitioning Algorithm based on Hybrid Hierarchical Clustering

BRDPHHC: A Balance RDF Data Partitioning Algorithm based on Hybrid Hierarchical Clustering 015 IEEE 17th Internatonal Conference on Hgh Performance Computng and Communcatons (HPCC), 015 IEEE 7th Internatonal Symposum on Cyberspace Safety and Securty (CSS), and 015 IEEE 1th Internatonal Conf

More information

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz Compler Desgn Sprng 2014 Regster Allocaton Sample Exercses and Solutons Prof. Pedro C. Dnz USC / Informaton Scences Insttute 4676 Admralty Way, Sute 1001 Marna del Rey, Calforna 90292 pedro@s.edu Regster

More information

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr)

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr) Helsnk Unversty Of Technology, Systems Analyss Laboratory Mat-2.08 Independent research projects n appled mathematcs (3 cr) "! #$&% Antt Laukkanen 506 R ajlaukka@cc.hut.f 2 Introducton...3 2 Multattrbute

More information

CS246: Mining Massive Datasets Jure Leskovec, Stanford University

CS246: Mining Massive Datasets Jure Leskovec, Stanford University CS46: Mnng Massve Datasets Jure Leskovec, Stanford Unversty http://cs46.stanford.edu /19/013 Jure Leskovec, Stanford CS46: Mnng Massve Datasets, http://cs46.stanford.edu Perceptron: y = sgn( x Ho to fnd

More information

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 An Iteratve Soluton Approach to Process Plant Layout usng Mxed

More information

Human Face Recognition Using Generalized. Kernel Fisher Discriminant

Human Face Recognition Using Generalized. Kernel Fisher Discriminant Human Face Recognton Usng Generalzed Kernel Fsher Dscrmnant ng-yu Sun,2 De-Shuang Huang Ln Guo. Insttute of Intellgent Machnes, Chnese Academy of Scences, P.O.ox 30, Hefe, Anhu, Chna. 2. Department of

More information

Meta-heuristics for Multidimensional Knapsack Problems

Meta-heuristics for Multidimensional Knapsack Problems 2012 4th Internatonal Conference on Computer Research and Development IPCSIT vol.39 (2012) (2012) IACSIT Press, Sngapore Meta-heurstcs for Multdmensonal Knapsack Problems Zhbao Man + Computer Scence Department,

More information

Available online at Available online at Advanced in Control Engineering and Information Science

Available online at   Available online at   Advanced in Control Engineering and Information Science Avalable onlne at wwwscencedrectcom Avalable onlne at wwwscencedrectcom Proceda Proceda Engneerng Engneerng 00 (2011) 15000 000 (2011) 1642 1646 Proceda Engneerng wwwelsevercom/locate/proceda Advanced

More information

Active Contours/Snakes

Active Contours/Snakes Actve Contours/Snakes Erkut Erdem Acknowledgement: The sldes are adapted from the sldes prepared by K. Grauman of Unversty of Texas at Austn Fttng: Edges vs. boundares Edges useful sgnal to ndcate occludng

More information

SHAPE RECOGNITION METHOD BASED ON THE k-nearest NEIGHBOR RULE

SHAPE RECOGNITION METHOD BASED ON THE k-nearest NEIGHBOR RULE SHAPE RECOGNITION METHOD BASED ON THE k-nearest NEIGHBOR RULE Dorna Purcaru Faculty of Automaton, Computers and Electroncs Unersty of Craoa 13 Al. I. Cuza Street, Craoa RO-1100 ROMANIA E-mal: dpurcaru@electroncs.uc.ro

More information

Deep Classification in Large-scale Text Hierarchies

Deep Classification in Large-scale Text Hierarchies Deep Classfcaton n Large-scale Text Herarches Gu-Rong Xue Dkan Xng Qang Yang 2 Yong Yu Dept. of Computer Scence and Engneerng Shangha Jao-Tong Unversty {grxue, dkxng, yyu}@apex.sjtu.edu.cn 2 Hong Kong

More information

Correlative features for the classification of textural images

Correlative features for the classification of textural images Correlatve features for the classfcaton of textural mages M A Turkova 1 and A V Gadel 1, 1 Samara Natonal Research Unversty, Moskovskoe Shosse 34, Samara, Russa, 443086 Image Processng Systems Insttute

More information

Smoothing Spline ANOVA for variable screening

Smoothing Spline ANOVA for variable screening Smoothng Splne ANOVA for varable screenng a useful tool for metamodels tranng and mult-objectve optmzaton L. Rcco, E. Rgon, A. Turco Outlne RSM Introducton Possble couplng Test case MOO MOO wth Game Theory

More information

An Optimal Algorithm for Prufer Codes *

An Optimal Algorithm for Prufer Codes * J. Software Engneerng & Applcatons, 2009, 2: 111-115 do:10.4236/jsea.2009.22016 Publshed Onlne July 2009 (www.scrp.org/journal/jsea) An Optmal Algorthm for Prufer Codes * Xaodong Wang 1, 2, Le Wang 3,

More information

Unsupervised Learning and Clustering

Unsupervised Learning and Clustering Unsupervsed Learnng and Clusterng Why consder unlabeled samples?. Collectng and labelng large set of samples s costly Gettng recorded speech s free, labelng s tme consumng 2. Classfer could be desgned

More information

Positive Semi-definite Programming Localization in Wireless Sensor Networks

Positive Semi-definite Programming Localization in Wireless Sensor Networks Postve Sem-defnte Programmng Localzaton n Wreless Sensor etworks Shengdong Xe 1,, Jn Wang, Aqun Hu 1, Yunl Gu, Jang Xu, 1 School of Informaton Scence and Engneerng, Southeast Unversty, 10096, anjng Computer

More information

Improving Web Image Search using Meta Re-rankers

Improving Web Image Search using Meta Re-rankers VOLUME-1, ISSUE-V (Aug-Sep 2013) IS NOW AVAILABLE AT: www.dcst.com Improvng Web Image Search usng Meta Re-rankers B.Kavtha 1, N. Suata 2 1 Department of Computer Scence and Engneerng, Chtanya Bharath Insttute

More information

Classification / Regression Support Vector Machines

Classification / Regression Support Vector Machines Classfcaton / Regresson Support Vector Machnes Jeff Howbert Introducton to Machne Learnng Wnter 04 Topcs SVM classfers for lnearly separable classes SVM classfers for non-lnearly separable classes SVM

More information

An enhanced representation of time series which allows fast and accurate classification, clustering and relevance feedback

An enhanced representation of time series which allows fast and accurate classification, clustering and relevance feedback An enhanced representaton of tme seres whch allows fast and accurate classfcaton, clusterng and relevance feedback Eamonn J. Keogh and Mchael J. Pazzan Department of Informaton and Computer Scence Unversty

More information

A New Approach For the Ranking of Fuzzy Sets With Different Heights

A New Approach For the Ranking of Fuzzy Sets With Different Heights New pproach For the ankng of Fuzzy Sets Wth Dfferent Heghts Pushpnder Sngh School of Mathematcs Computer pplcatons Thapar Unversty, Patala-7 00 Inda pushpndersnl@gmalcom STCT ankng of fuzzy sets plays

More information

Spam Filtering Based on Support Vector Machines with Taguchi Method for Parameter Selection

Spam Filtering Based on Support Vector Machines with Taguchi Method for Parameter Selection E-mal Spam Flterng Based on Support Vector Machnes wth Taguch Method for Parameter Selecton We-Chh Hsu, Tsan-Yng Yu E-mal Spam Flterng Based on Support Vector Machnes wth Taguch Method for Parameter Selecton

More information

Three supervised learning methods on pen digits character recognition dataset

Three supervised learning methods on pen digits character recognition dataset Three supervsed learnng methods on pen dgts character recognton dataset Chrs Flezach Department of Computer Scence and Engneerng Unversty of Calforna, San Dego San Dego, CA 92093 cflezac@cs.ucsd.edu Satoru

More information

Sequential search. Building Java Programs Chapter 13. Sequential search. Sequential search

Sequential search. Building Java Programs Chapter 13. Sequential search. Sequential search Sequental search Buldng Java Programs Chapter 13 Searchng and Sortng sequental search: Locates a target value n an array/lst by examnng each element from start to fnsh. How many elements wll t need to

More information

An Entropy-Based Approach to Integrated Information Needs Assessment

An Entropy-Based Approach to Integrated Information Needs Assessment Dstrbuton Statement A: Approved for publc release; dstrbuton s unlmted. An Entropy-Based Approach to ntegrated nformaton Needs Assessment June 8, 2004 Wllam J. Farrell Lockheed Martn Advanced Technology

More information

Hierarchical clustering for gene expression data analysis

Hierarchical clustering for gene expression data analysis Herarchcal clusterng for gene expresson data analyss Gorgo Valentn e-mal: valentn@ds.unm.t Clusterng of Mcroarray Data. Clusterng of gene expresson profles (rows) => dscovery of co-regulated and functonally

More information

Using Neural Networks and Support Vector Machines in Data Mining

Using Neural Networks and Support Vector Machines in Data Mining Usng eural etworks and Support Vector Machnes n Data Mnng RICHARD A. WASIOWSKI Computer Scence Department Calforna State Unversty Domnguez Hlls Carson, CA 90747 USA Abstract: - Multvarate data analyss

More information

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints Australan Journal of Basc and Appled Scences, 2(4): 1204-1208, 2008 ISSN 1991-8178 Sum of Lnear and Fractonal Multobjectve Programmng Problem under Fuzzy Rules Constrants 1 2 Sanjay Jan and Kalash Lachhwan

More information

Performance Evaluation of Information Retrieval Systems

Performance Evaluation of Information Retrieval Systems Why System Evaluaton? Performance Evaluaton of Informaton Retreval Systems Many sldes n ths secton are adapted from Prof. Joydeep Ghosh (UT ECE) who n turn adapted them from Prof. Dk Lee (Unv. of Scence

More information

An AAM-based Face Shape Classification Method Used for Facial Expression Recognition

An AAM-based Face Shape Classification Method Used for Facial Expression Recognition Internatonal Journal of Research n Engneerng and Technology (IJRET) Vol. 2, No. 4, 23 ISSN 2277 4378 An AAM-based Face Shape Classfcaton Method Used for Facal Expresson Recognton Lunng. L, Jaehyun So,

More information

Wishing you all a Total Quality New Year!

Wishing you all a Total Quality New Year! Total Qualty Management and Sx Sgma Post Graduate Program 214-15 Sesson 4 Vnay Kumar Kalakband Assstant Professor Operatons & Systems Area 1 Wshng you all a Total Qualty New Year! Hope you acheve Sx sgma

More information

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces Range mages For many structured lght scanners, the range data forms a hghly regular pattern known as a range mage. he samplng pattern s determned by the specfc scanner. Range mage regstraton 1 Examples

More information

A Novel Term_Class Relevance Measure for Text Categorization

A Novel Term_Class Relevance Measure for Text Categorization A Novel Term_Class Relevance Measure for Text Categorzaton D S Guru, Mahamad Suhl Department of Studes n Computer Scence, Unversty of Mysore, Mysore, Inda Abstract: In ths paper, we ntroduce a new measure

More information

Related-Mode Attacks on CTR Encryption Mode

Related-Mode Attacks on CTR Encryption Mode Internatonal Journal of Network Securty, Vol.4, No.3, PP.282 287, May 2007 282 Related-Mode Attacks on CTR Encrypton Mode Dayn Wang, Dongda Ln, and Wenlng Wu (Correspondng author: Dayn Wang) Key Laboratory

More information

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur FEATURE EXTRACTION Dr. K.Vjayarekha Assocate Dean School of Electrcal and Electroncs Engneerng SASTRA Unversty, Thanjavur613 41 Jont Intatve of IITs and IISc Funded by MHRD Page 1 of 8 Table of Contents

More information

Feature Reduction and Selection

Feature Reduction and Selection Feature Reducton and Selecton Dr. Shuang LIANG School of Software Engneerng TongJ Unversty Fall, 2012 Today s Topcs Introducton Problems of Dmensonalty Feature Reducton Statstc methods Prncpal Components

More information