Reinforcement Learning-based Feature Selection For Developing Pedagogically Effective Tutorial Dialogue Tactics
|
|
- Oscar Perkins
- 6 years ago
- Views:
Transcription
1 Renforcement Learnng-based Feature Selecton For Developng Pedagogcally Effectve Tutoral Dalogue Tactcs 1 Introducton Mn Ch, Pamela Jordan, Kurt VanLehn, Moses Hall {mc31, pjordan+, vanlehn+, mosesh}@ptt.edu Learnng Research Development Center & Intellgent System Program, Unversty of Pttsburgh Abstract. Gven the subtlety of tutoral tactcs, dentfyng effectve pedagogcal tactcal rules from human tutorng dalogues and mplementng them for dalogue tutorng systems s not trval. In ths work, we used renforcement learnng (RL to automatcally derve pedagogcal tutorng dalog tactcs. Past research has shown that the choce of the features sgnfcantly affects the effectveness of the learned tactcs. We defned a total of 18 features whch we classfed nto four types. Frst, we compared fve feature selecton methods and overall upper-bound method seems to be most effcent. Then we compared the four types of features and found that temporal stuaton and autonomy related features are sgnfcantly more relevant and effectve to tutoral decsons than ether performance or stuaton related features. One challengng ssue confrontng dalogue tutorng systems s how to select the best tutorng actons for any student state so as to maxmze learnng gans. We call ths decson makng tutorng tactcs snce t generally governs bref epsodes of tutorng dalogue, such as a sngle step, and seems to be crucal for achevng further mprovements n pedagogcal effectveness. Bad tutoral decsons have the potental to bore or frustrate students to the pont that they fal to learn. Many exstng dalogue systems use hand-crafted tactcal rules derved from analyzng human dalogues [6][13]. However, dentfyng and mplementng these tactcal rules s not trval. It has been shown that expert human tutors generally employ a range of tutorng actons and ther decsons depend n subtle ways on the student s competence, student s self-confdence and other factors. For example, one mportant tutorng tactc s whether the tutor should tell the student the answer drectly for a step or whether he should elct the step from the student wth a prompt or a seres of questons. Collns et al. suggest that when a student s unfamlar wth the target knowledge, the tutor should tell t drectly; when a student becomes more famlar wth the knowledge, the tutor should elct the answer va questons; and once a student has mastered the knowledge, t doesn't matter whether t s told or elcted [4][5]. Most exstng dalogue tutorng systems wth hand-crafted rules react sensbly when students exhbt less than optmal behavors; however, ther tutoral nteractons are much stffer than those of human tutors and they have not yet acheved the same effectveness that one-on-one, face-toface expert human tutors have [1]. In recent years, work n desgnng spoken dalogue systems has proposed several data-drven methodologes; one of them s Renforcement Learnng (RL [3][12][13]. It has been shown that RL s effectve at automatcally learnng the best acton to take at any state n a dalogue. And such success nspred us to use RL for dalogue tutorng systems. Our work can be dvded nto three stages. In the frst stage, we bult an ntal tutoral dalogue system and collected an exploratory corpus by usng the system to tran 64 students. In ths ntal system, decsons on certan types of tutorng actons were made randomly. In the second stage, we used RL on the exploratory corpus to derve tutorng tactcs for these actons and then ncorporated the learned tactcs nto the ntal system. The modfed system s dentcal to the
2 ntal one except that the tutorng actons that were prevously made randomly are now based on the learned tutorng tactcs. In the last stage, the modfed system s beng used to tran a new group of students so that we compare ther learnng performance to that of the students prevously traned on the ntal system. We expect the modfed system to be so well attuned to students that t wll be more effectve than the ntal system. One tactcal decson nvestgated n ths study was whether the tutor should tell the next step to the student or elct t from the student. We defned 18 features to model the dalogue and students state. They were selected based upon the four types of features that were shown n [7] to be relevant for human tutors to makng ther tutoral decsons: autonomy, temporal stuaton, stuaton, and performance respectvely. Among 18 features, features 1-3 are autonomy related based upon the amount of the work that the tutor has let students do; features 4-8 cover tmerelated nformaton such as tme spent on the tranng so far; features 9-11 are stuaton related ncludng whether the student and tutor are now problem solvng or n post dscusson, the dffculty level of the problem and so on; and fnally, features cover the performance nformaton such as the correctness of student s prevous answers and the ablty of student. In an RL model, the sze of the state space ncreases exponentally as the number of nvolved features ncreases. In order to learn effectve tutorng tactcs, we should have a corpus that covers each of these states at least once, whch means 2 18 n our case. However, t s almost mpossble to do so gven the hgh cost of collectng educatonal data. On the other hand, the learned polcy may become too subtle to be necessary. Fgure 1 shows an example of a learned polcy nvolvng fve features: [1, 2, 4, 15, 16], n whch feature 2 s an autonomy feature defned as the percentage of elctaton the students have receved so far. Each of the fve features was converted from a real number to a bnary value based upon ther medan scores, for example, f2='.4982' means f feature 2 value s above.4982, t s 1 otherwse, t s 0. There were a total of 32 rules learned: n 10 stuatons the tutor should elct, n 19 t should tell; n the remanng 3 cases ether wll do. For example, when all of the features are 0 then the tutor should tell as 0:0:0:0:0 the frst n the lst of tells. As you can see, fve features provde a large space s already qute complex and s already much more subtle than most of the tutoral tactcs derved from analzng human turoral dalogues [4][5]. KC22 'features'=[1, 2, 4, 15,16], 'cutoff'=[f1= ' f2='.4982' f4=' ' f15='.6154' f='.2683 ], 'polcy': 'elct: [0:0:0:0:1, 0:0:0:1:0, 1:0:0:0:0, 1:0:0:0:1, 1:0:0:1:0, 1:1:0:1:1, 0:0:1:0:0, 0:0:1:0:1, 1:0:1:0:0, 1:0:1:0:1], 'tell: [0:0:0:0:0, 0:0:0:1:1, 0:1:0:0:0, 0:1:0:0:1, 0:1:0:1:0, 0:1:0:1:1, 1:1:0:0:1, 1:1:0:1:0, 0:0:1:1:0, 0:0:1:1:1, 0:1:1:0:0, 0:1:1:0:1, 0:1:1:1:0, 0:1:1:1:1, 1:0:1:1:1, 1:1:1:0:0, 1:1:1:0:1, 1:1:1:1:0, 1:1:1:1:1] 'else: [1:0:0:1:1, 1:1:0:0:0, 1:0:1:1:0] Fgure 1. A Learned Polcy Based On Fve Features On KC22 Based on the sze of the exploratory corpus we collected n the frst stage, we decded that no more than sx features should be used. Prevous research nvolvng RL and feature selecton n dalogue systems ether focused on selectng features wth certan characterstcs [2] [7][8] or nvestgated a relatvely small number of features [10][11]. Therefore, n ths study, we proposed four general RL-based feature selecton methods.
3 2 Background Past research on usng RL to mprove spoken dalogue systems has commonly used Markov Decson Processes (MDP s. An MDP [8][9] descrbes a stochastc control process and formally corresponds to a 4-tuple (S,A,T,R, where S = {s } =1..m, s a fnte set of process states; A={a k } k=1..m s a fnte set of actons; T : S A S [0, 1] s a set of transton probabltes between states that descrbe the dynamcs of the modeled system; for example: P t (s s j, a k s the probablty that the model would transton from state s j to state s by takng acton a k at tme t; and R : S A S R denotes a reward model that assgns rewards to state transtons and models payoffs assocated wth such transtons. The goal of usng MDPs s to determne the best polcy π*: the set of actons the model should take at each state s to maxmze ts expected cumulatve utlty (V-value, whch can be calculated from the followng recursve equaton: V( s = Ps s j π*( s s j [ R π*( s ss j + γv( s As long as a proper state space, acton space and reward functon are set up, an MDP allows one to automatcally derve and compute the best polcy. For dalogue tutorng systems, dervng effectve tutoral tactcs from tutorng dalogues can be naturally cast n the MDP formalsm: for the state space, each s can be vewed as a vector of features representng the tutorng dalogue context, the student s knowledge level and so on. Acton space corresponds to tutorng actons, e.g. elct or tell; the reward functon corresponds to students learnng gans. For tutorng dalogues, the reward s a delayed reward because for each state the reward s not known untl students have completed the tutorng and taken the post-test. In ths work, we used an MDP package that was desgned for Tetreault & Ltman s studes snce t has proven to be both relable and successful [10][11]. In prevous studes, Tetreault & Ltman prmarly nvestgated methods for evaluatng whether certan features would mprove polcy effectveness. There are two man dfferences between our study and thers: frstly, they used a prevously collected corpus that was not exploratory wth respect to tutoral tactcs to tran an MDP model n that the tutor often used only one type of acton n many dalogue states, whch severely lmted the types of questons that they could nvestgate[10][11]; whle we use a more exploratory corpus by tranng students on a dalogue tutorng system n whch multple actons can often be taken and random tactcal decsons were made. Thus t s better suted for creatng an MDP. Secondly, they dd not need to address the problem of general feature selecton methods snce they only used fve features whle we had to select up to sx of 18. To evaluate the learned polces, we use three crtera: Expected Cumulatve Reward (ECR and lower-bound and hgher-bound of the 95% confdence nterval. ECR s the average reward one would expect n the MDP and t s calculated by normalzng the V-value of each state by the number of tmes t occurs as a start state n a dalogue and then summng over all states [11]. The confdence nterval represents how relable the ECR of a learned polcy s [10]. The wder t s, the less relable the polcy s ECR s. For example, for a learned polcy A derved from feature 2 alone on defnton of sprng potental energy (KC22 s f the percent of elct so far s less than 49.82%, tutor should elct otherwse the tutor should tell. ; A has ECR = 3.02 (range [-100, 100] wth a 95% confdence nterval= [- 2.71, 8.45], whch means there s a 95% chance that the ECR of the learned polcy s between a lower-bound of and an upper-bound of In Fgure 1, also on KC22 but nvolves fve features: 1, 2, 4, 15 and 16. It has ECR = wth a 95% confdence nterval= [23.49, 50.51], whch s much more effectve than polcy A because even ts lower-bound s much hgher than polcy A s upper-bound. j ]
4 Sometmes we encounter stuatons n whch the ECR for A s the same as for B; but the confdence nterval of A s much narrower than that of B. By only usng the three crtera descrbed above, polcy A and B cannot be compared. Therefore, we defne the hedge of a learned polcy as: ECR Hedge = ({ upper bound} { lower bound} By usng the hedge, we can say that polcy A s better than polcy B snce the hedge of A s hgher than the hedge of polcy B. 3 Experment The doman chosen for ths study s work and energy covered n college physcs. The procedure was as follows students: (1 read a short textbook, whch descrbes the major prncples and concepts; (2 take a pre-test; (3 work through seven open-ended tranng problems wth a dalogue tutorng system (Cordllera [14]; and (4 take a post-test that s dentcal to the pre-test. In the stage one, 64 college students who had not taken any college physcs completed the experment, recevng payment for ther partcpaton. Students needed 8-15 hours to complete the procedures and usually requred 4-6 sessons of about 2 hours each. Thus, the collected corpus comprses 64 human-computer tutorng dalogues and each dalogue s an extended nteracton wth one student that covers seven dfferent college-level physcs problems. The number of stateacton pars n each of the 64 collected tutoral dalogues vares from 700 to 900. All tranng problems, and all pre- and post-tests problems were selected from 127 quanttatve and qualtatve problems collected from varous physcs lterature. In order to solve these 127 problems, 32 unque knowledge components (KCs were dentfed as necessary. A KC s a generalzaton of everyday terms lke concept, prncple, fact, or skll, and cogntve scence terms lke schema, producton rule, msconcepton, or facet [14]. For example, KC22 1 s both a concept and a prncple whle KC23 2 s a fact. KC22 conssts of both procedural knowledge and declaratve knowledge whle KC23 s manly declaratve knowledge and thus learnng these two KCs clearly nvolves dfferent cogntve sklls. Therefore, for dfferent KCs, we expect that dfferent features should be consdered when makng a tactcal decson and dfferent tactcal decsons should be derved. In order to learn KC-specfc tutoral tactcs, students pre- and post-test scores were also calculated per KC. It turned out our tranng, pre- and post-problems covered 27 of the 32 possble KCs. Three of the 27 KCs showed up n the tutorng dalogue but were not assocated wth any acton decson ponts; fve KCs concded only once or twce wth decson ponts; and one KC dd not appear n the pre- and post-tests. Therefore, we were left wth 18 KCs for whch t s possble for us to derve tutorng tactcs. Comparsons of pre- and post-test scores ndcated that students dd learn durng ther tranng wth Cordllera: ther post-test scores were sgnfcantly hgher than ther pre-test scores: t(126= 3.36, p= Smlar to Tetreault & Ltman s work, we used students normalzed learnng gans (NLGs for each KC as the delayed reward functon, whch s defned as: Posttest( KC Pr etest( KC NLG( KC =. 1 Pr etest( KC 1 If an object and a sprng are n a system, then ther sprng potental energy s 0.5* k*d^2, where k s the sprng constant and d s the dsplacement of the object relatve to the equlbrum poston of the sprng. 2 The unt for energy s the Joule.
5 For each KC, the 64 students were splt nto two groups by the medan value of ther NLG on t, so the better-performng students dalogues were gven a postve reward of +100, whle the other half were assgned a negatve reward of The rewards were assgned n the fnal dalogue state. Addtonally, we annotated our tutorng dalogues and acton decsons based on whch KCs a tutor acton or tutor-student par of turns covered; Table 1 shows an example. Here lnes 1, 5, 6, 7, 9 and 11 are tutor turns, the acton type and ther target KCs are lsted on the rght. For example, n the frst utterance s an elct acton regardng KC22 whle the ffth s a tell acton drectly statng KC22. Lnes 2 and 12 are student responses and they are both correct (as ndcated by the + ; row 2 s about KC22 whle row 12 s about KC27. Table 1. Examples of Tutoral Dalogue wth KC Labels 1. T: At T2 we also know the sprng s uncompressed. So, what type of energy elct, can we easly fnd now? KC22 2. S: spe +KC22 3. T: Yes, I agree. 4. T: At ths pont we know the values for KE2 and GPE2. It s tme to compute the value for SPE2. 5. T: It s mportant to remember that we cannot fnd the SPE of an OBJECT. We tell, KC22 need to fnd the sprng potental energy of a par: object-sprng. 6. T: It s possble to compute the sprng potental energy at T2 because we know tell the sprng s uncompressed. That s, d2=0 m. 7. T: Usng ths nformaton we can nfer that the sprng potental energy of the tell, KC22 block-sprng par at T2 s SPE2=0 J. KC23 8. T: Now we have all we need to 9. T: TME of the system at T2 s the sum of the block`s KE and the block-earth par`s GPE because the value for the sprng potental energy s zero. In ths case TME2=11.88 J+1.62 J =13.5 J. 10. T: At ths pont we have found the value for TME2. We can now easly fnd the system`s TME at all other tme ponts snce the system s solated. 11. T: What prncple gves us the system`s total mechancal energy at every tme pont, knowng ts value at T2 and also gven that the system s solated? tell, KC23 KC24 elct, KC S: student conservaton of energy + KC T: Yes. 4 Methods The four feature selecton approaches we propose n ths paper are farly straghtforward and share the same procedure, whch conssts of the three phases descrbed below. Phase 1: For each of 18 features n the state space, use MDP to get a snglefeature-polcy. Phase 2: Sort all of the features based on the learned sngle-feature-polces from hgh to low by one of the followng measures: ECR Lower-bound Upper-bound Hedge Phase 3: Start wth the frst feature n the sorted feature lst, add one feature at a tme to the MDP and learn a new polcy. Repeat ths process repeat 5 tmes.
6 Based on the sortng crtera n phase 2, we named our four feature selecton methods: ECR, lower-bound, upper-bound, and hedge respectvely. Snce these values were calculated from the sngle-feature-polces learned usng MDP n phase 1, the four methods are RL-based. We expect them to be more effectve than a random selecton. To test our expectatons, we created 120 random polces by runnng the MDP package on randomly selected features (20 rounds for each randomly selected feature set, where each set contaned between one and sx features. Therefore, for each KC and each feature set sze, we learned one tutorng tactc for each RL-based method plus 20 for random feature selecton. Ths gave us (1*4+20*6 =144 polces. For each KC, we selected one polcy that had the hghest ECR, lower-bound and upper-bound from the 144 learned polces and named t the best polcy. To quanttatvely evaluate how much less effectve a learned tutorng tactc s than the best polcy, we defned a normalzed ECR (NECR for a learned tutorng tactc as: NECR(KC, N, Method Max _ ECR( KC = Mn _ ECR( KC k j 1 ECR(KC, N, Method k - Mn_ECR(KC = C Max_ECR(KC - Mn_ECR(KC = C max N {1..6}, m { all methods} mn N {1..6}, m { all methods} ( ECR( KC, N, Method ( ECR( KC, N, Method Max_ECR(KC and Mn_ECR(KC are the maxmum and mnmum ECR among all 144 of the learned polces for KC. C s a constant wth C=1 for each of the four RL-based methods and C=20 for random feature selecton. The maxmum NECR for a learned polcy s 1 f the learned polcy s the best one and the mnmum s 0 f the learned polcy s the worst one. 5 Results In order to evaluate the feature selecton methods and how the feature effectveness, we have two man goals. Frst, we wll compare the four RL-based feature selecton methods aganst random feature selecton; second we wll nvestgate whch features seem to be more mportant for dervng the best tutorng tactcs for decdng to elct/tell across all KCs. 5.1 Comparng four RL-based methods aganst random feature selecton Table 2. Comparng the Average NECR of Fve Selecton Methods for Increasng Feature Set Szes Number Features Involved Upper-bound ECR Hedge Lower-bound Random m m NECR s defned as how much a learned tutorng tactcs s less effectve than the best polcy. Table 2 shows the average NECR gven the number of selected features and feature selecton
7 method across all 18 KCs. As expected, random feature selecton has the worst average NECR regardless of the number of features nvolved. Overall, f the number of features 3, then the ECR approach s the best feature selecton method; but f the number of features s between 4 and 6, then the upper-bound method s the best. As the number of features nvolved ncreases, the effectveness of the learned polces for elct/tell tends to become better, except for upper-bound, whch has a better average NECR when there are fve features than when there are sx. Overall, usng the upper-bound method to select fve features seems to be the most effectve method across all KCs on elct/tell acton decson. On the other hand, across all 18 KCs the number of tmes that each method found the best polcy s: fourteen tmes for random, three for upper-bound, two for ECR, one for hedge and none for lower-bound method. The total dd not add up to 18 KCs because for KC1, three methods, ecr, hedge, and upper-bound all found the same best polcy. Therefore, although random feature selecton has the worse average NECR than other four RL-based methods, t s an effectve way to fnd best polces. However, note that the random feature selecton method was repeated 120 tmes for each KC and so t has a total of 120*18=2160 chances to fnd the 14 best polces; whle the upper-bound method was appled only 6 tmes for each KC and thus has a total of 108 chances. Addtonally, because our state space s stll relatvely small, we expect that the performance of random selecton would decrease sgnfcantly as the number of features n the state space ncreases. However, for the four RL-based feature selecton methods, ncreasng the number of features would not decrease ther effectveness snce they do not drectly depend on the number of features n the state space. Moreover we compared the best polcy learned for each KC by the upper-bound method and those learned by random selecton. Over all 18 KCs, the upper-bound polces wth just 108 attempts were only 9.46% less effectve than random feature selecton polces wth 2160 attempts. These results ndcate that that n educaton, features that may result hgher learnng gans should always be always be consdered by the tutor when makng decsons. Ths s lkely due to the fact that n the worst case a student wll smply not learn rather than lose nformaton so the cost of ncorporatng superfluous feature s low Frequency of Features n Best Polces Fgure 2. The Frequency of Each Feature Shown In Best Polces. Fgure 2 shows the frequency wth whch each feature appears n the best polces. Ths frequency dffers sgnfcantly among the four feature types: F(3= 7.47, p= There s no sgnfcant dfference between the three autonomy and fve temporal stuaton related features. When combned they are sgnfcantly more frequent than ether the performance or stuaton related features: t(12= 2.74, p=0.18 and t(10= 4.26, p=0.002 respectvely. Consstent wth prevous research based on analyzng human tutoral dalogue, autonomy related features seemed to be more relevant n dervng effectve tutoral tactcs. Addtonally, we found that temporal
8 stuaton related features were also relevant, even more so than the performance related ones when decdng whether to elct or tell. Ths was not ndcated n prevous lterature on human tutoral dalog analyss. One possble explanaton for ths was that n most of the pror lterature, temporal stuaton related factors were often not consdered. 6 Conclusons and Future Work In ths paper, we descrbed our work on applyng RL methods to derve effectve tutorng tactcs for elct/tell. We showed that dervng effectve tutorng tactcs from tutorng dalogues can be cast n the MDP formalsm. Addtonally, we proposed four RL-based doman-general feature selecton methods. We found the upper-bound method to be more effectve than the others. One of our goals for future work s to nvestgate how to stll get reasonable polces wthout annotatng ndvdual KCs n the dalogues. The annotaton process s prohbtvely tmeconsumng and t s not unusual for doman experts to dsagree [14]. Another of our goals for future work s to determne how to learn one reasonable polcy for all KCs wthout sacrfcng too much of the expected effectveness. Another two mportant ssues are how to avod the expensve ntal data collecton and how to combne the new data wth the exstng data so that we can learn even more powerful polces. Acknowledgments We would lke to thank the NLT group and Dane J. Ltman for ther comments. Support for ths research was provded by NSF grants # References [1] B. S. Bloom, The 2 sgma problem: The search for methods of group nstructon as effectve as oneto-one tutorng., Educatonal Researcher,, no. 13, pp. 4 16, [2] M. Frampton and O. Lemon Renforcement learnng of dalogue strateges usng the user's last dalogue act. In IJCAI Wkshp. on K&R n Practcal Dalogue Systems. [3] J. Henderson, O. Lemon, and K. Georgla Hybrd renforcement/supervsed learnng for dalogue polces from communcator data. In IJCAI Wkshp. on K&R n Practcal Dalogue Systems. [4] Collns, A., and Stevens, A Goals and Strateges of Inqury Teachers: In R. Vlaser (Ed Advances n Instructonal Psychology (vol. 2 Hllsdale, NJ: Erlbaum Assoc [5] Collns, A. (1996. Desgn ssues for learnng envronments. In S. Vosnadou, E. D. Corte, R. Glaser, & H. Mandl (Eds., Internatonal perspectves on the desgn of technology-supported learnng envronments (pp Mahwah, NJ: Erlbaum, (Bloom, [6] M. Evens and J. Mchael, One-on-One Tutorng by Humans and Computers, Lawrence Erlbaum Assocates, Inc., [7] Johanna D. Moore, Kaska Porayska-Pomsta, Sebastan Varges, Claus Znn: Generatng Tutoral Feedback wth Affect. FLAIRS Conference 2004 [8] S. Sngh, M. Kearns, D. Ltman, and M. Walker Renforcement learnng for spoken dalogue systems. In Proc. NIPS 99. [9] R. Sutton and A. Barto Renforcement Learnng. The MIT Press. [10] J. Tetreault, D. Bohus, and D. Ltman, Estmatng the relablty of mdp polces: a confdence nterval ap proach, n NAACL, [11] J. Tetreault and D. Ltman Comparng the utlty of state features n spoken dalogue usng renforcement learnng. In NAACL. [12] J. Wllams, P. Poupart, and S. Young. 2005a. Factored partally observable markov decson processes for dalogue management. In IJCAI Wkshp. on K&R n Practcal Dalogue Systems. [13] M. Walker An applcaton of renforcement learnng to dalogue strategy selecton n a spoken dalogue system for emal. JAIR, 12. [14] VanLehn, K, Jordan, & Dane Developng pedagogcally effectve tutoral dalogue tactcs: Experments and a testbed. In proceedngs of SLaTE Workshop.
CSCI 104 Sorting Algorithms. Mark Redekopp David Kempe
1 CSCI 104 Sortng Algorthms Mark Redekopp Davd Kempe 2 Algorthm Effcency SORTING 3 Sortng If we have an unordered lst, sequental search becomes our only choce If we wll perform a lot of searches t may
More informationTEACHING SPEAKING USING THE INFORMATION GAP TECHNIQUE. By Dewi Sartika * ABSTRACT
TEACHING SPEAKING USING THE INFORMATION GAP TECHNIQUE By Dew Sartka * Unversty of Syah Kuala, Banda Aceh ABSTRACT Ths study was amed at fndng out f there would be a sgnfcant dfference n achevement between
More informationIdentifying Intention Posts in Discussion Forums
Identfyng Intenton Posts n Dscusson Forums Zhyuan Chen, Bng Lu Department of Computer Scence Unversty of Illnos at Chcago Chcago, IL 60607, USA czyuanacm@gmal.com,lub@cs.uc.edu Mechun Hsu, Malu Castellanos,
More informationFactors Affecting Students' Performance. 16 July 2006
Factors Affectng Students' Performance Nasr Harb 1 * Department of Economcs College of Busness & Economcs Unted Arab Emrates Unversty P.O. Box 17555 Al-An, UAE Tel.: 971 3 7133228 Fax: 971 3 7624384 E-mal:
More informationSemantic Inference at the Lexical-Syntactic Level
Semantc Inference at the Lexcal-Syntactc Level Roy Bar-Ham and Ido Dagan Computer Scence Department Bar-Ilan Unversty Ramat-Gan 52900, Israel {barhar, dagan}@cs.bu.ac.l Iddo Greental Lngustcs Department
More informationSemantic Inference at the Lexical-Syntactic Level
Semantc Inference at the Lexcal-Syntactc Level Roy Bar-Ham and Ido Dagan Computer Scence Department Bar-Ilan Unversty Ramat-Gan 52900, Israel {barhar, dagan}@cs.bu.ac.l Iddo Greental Lngustcs Department
More informationAvailable online at Procedia Economics and Finance 2 ( 2012 )
Avalable onlne at www.scencedrect.com Proceda Economcs and Fnance 2 ( 2012 ) 353 362 2nd Annual Internatonal Conference on Qualtatve and Quanttatve Economcs Research Abstract (QQE 2012) A Survey of Tha
More informationThe Differential in Earnings Premia Between Academically and Vocationally Trained Males in the United Kingdom
ISSN 2045-6557 The Dfferental n Earnngs Prema Between Academcally and Vocatonally Traned Males n the Unted Kngdom Gavan Conlon June 2001 Publshed by Centre for the Economcs of Educaton London School of
More informationNon-Profit Academic Project, developed under the Open Acces Initiative
Red de Revstas Centífcas de Amérca Latna, el Carbe, España y Portugal Sstema de Informacón Centífca Eduardo Islas, Mguel Pérez, Gullermo Rodrguez, Israel Paredes, Ivonne Ávla, Mguel Mendoza E-learnng Tools
More informationIn ths paper we want to show that the possble analyses of ths problem wthn the framework of PSG are lmted by combnatons of the followng basc assumpton
On Non-ead Non-Movement In: G. Goerz (ed.): KONVNS 9, Sprnger Verlag, 99, pp 8- Klaus Netter eutsches Forschungszentrum fur Kunstlche Intellgenz Gmb Stuhlsatzenhausweg, -00 Saarbrucken, Germany e-mal:
More informationQUERY TRANSLATION FOR CROSS-LANGUAGE INFORMATION RETRIEVAL BY PARSING CONSTRAINT SYNCHRONOUS GRAMMAR
QUERY TRANSLATION FOR CROSS-LANGUAGE INFORMATION RETRIEVAL BY PARSING CONSTRAINT SYNCHRONOUS GRAMMAR FRANCISCO OLIVEIRA 1, FAI WONG 1, KA-SENG LEONG 1, CHIO-KIN TONG 1, MING-CHUI DONG 1 1 Faculty of Scence
More informationScenario Development Approach to Management Simulation Games
do: 10.1515/tms-2014-0022 2014 / 17 Scenaro Development Approach to Management Smulaton Games Jana Bkovska, Rga Techncal Unversty Abstract The paper ntroduces a scenaro development approach to management
More informationEfficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects
Effcent Estmaton of Tme-Invarant and Rarely Changng Varables n Fnte Sample Panel Analyses wth Unt Fxed Effects Thomas Plümper and Vera E. Troeger Date: 24.08.2006 Verson: trc_80 Unversty of Essex Department
More informationCultural Shift or Linguistic Drift? Comparing Two Computational Measures of Semantic Change
Cultural Shft or Lngustc Drft? Comparng Two Computatonal Measures of Semantc Change Wllam L. Hamlton, Jure Leskovec, Dan Jurafsky Department of Computer Scence, Stanford Unversty, Stanford CA, 94305 wlef,jure,jurafsky@stanford.edu
More informationAlignment USING COURSE, PROGRAM, AND INSTITUTIONAL CURRICULAR MAPS FOR ALIGNMENT
Algnment USNG COURSE PROGRAM AND NSTTUTONAL CURRCULAR MAPS FOR ALGNMENT "Mappng" s ntentonal1y desgned to capture or nvestgate currcular coherence by eplorng the algnment between learnng outcomes courses
More informationImprovement of Text Dependent Speaker Identification System Using Neuro-Genetic Hybrid Algorithm in Office Environmental Conditions
IJCSI Internatonal Journal of Computer Scence Issues, Vol. 1, 2009 ISSN (Onlne): 1694-0784 ISSN (Prnt): 1694-0814 42 Improvement of Text Dependent Speaker Identfcaton System Usng Neuro-Genetc Hybrd Algorthm
More informationLong Distance Wh-movement in Seereer: Implications for Intermediate Movement
Long Dstance Wh-movement n Seereer: Implcatons for Intermedate Movement Nco Baer U Berkeley nbbaer@berkeley.edu PL 38 March 29, 2014 1 Introducton Queston: What motvates ntermedate movement n a successve-cyclc
More informationA Training Manual for Educators K16
A Tranng Manual for Educators K16 ArzonA 2010 Servce for a Lfe Tme A Tranng Manual for Educators K16 Edted by Debb Bertolet, Joan Brd, and Sar Nms Funded by State Farm Learn and Serve Arzona Mesa Publc
More informationintellect edison.dadeschools.net i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i College board academy
Edson where nventon begns mathematcs scence languages socal studes dual enrollment electves ntellect College board academy The College Board Academy offers an array of hgh-level courses for students who
More informationvision values& Values & Vision NEVADA ARTS COUNCIL STRATEGIC PLAN A society is reflected by the state of its arts. All of Nevada deserves
values& vson A socety s reflected by the state of ts arts. All of Nevada deserves access to the arts. NEVADA ARTS COUNCIL STRATEGIC PLAN 2004-2007 Values & Vson The Nevada Arts Councl s a dvson of the
More information1 st HALF YEARLY MONITORING REPORT OF (JAMIA MILLIA ISLAMIA) on MDM for the State of UTTAR PRADESH
Annexure V 1 st HALF YEARLY MONITORING REPORT OF (JAMIA MILLIA ISLAMIA) on MDM for the State of UTTAR PRADESH Perod: 1 st Aprl 2013 to 30 st September 2013 Dstrcts Covered 1. BARABANKI 2. LUCKNOW 3. SANT
More informationA MULTIOBJECTIVE OPTIMIZATION FOR THE EWMA AND MEWMA QUALITY CONTROL CHARTS
Invese Poblems, Desgn and Optmzaton Symposum Ro de Janeo, Bazl, 24 A MULTIOBJECTIVE OPTIMIZATION FOR THE EWMA AND MEWMA QUALITY CONTROL CHARTS Fancsco Apas Depatamento de Estadístca e Investgacón Opeatva
More informationLecture 10: Reinforcement Learning
Lecture 1: Reinforcement Learning Cognitive Systems II - Machine Learning SS 25 Part III: Learning Programs and Strategies Q Learning, Dynamic Programming Lecture 1: Reinforcement Learning p. Motivation
More informationReinforcement Learning by Comparing Immediate Reward
Reinforcement Learning by Comparing Immediate Reward Punit Pandey DeepshikhaPandey Dr. Shishir Kumar Abstract This paper introduces an approach to Reinforcement Learning Algorithm by comparing their immediate
More informationPatterson, Carter see new county jail in the future
, k» \ f Patterson, Carter see new county jal n the future SHERIFF PATTERSON Because of leadershp post By NORRIS R. MCDOWELL A new jal for Clnton County? The possblty has been rased by a recent letter
More informationAdaptive Generation in Dialogue Systems Using Dynamic User Modeling
Adaptive Generation in Dialogue Systems Using Dynamic User Modeling Srinivasan Janarthanam Heriot-Watt University Oliver Lemon Heriot-Watt University We address the problem of dynamically modeling and
More informationWhat s in a Step? Toward General, Abstract Representations of Tutoring System Log Data
What s in a Step? Toward General, Abstract Representations of Tutoring System Log Data Kurt VanLehn 1, Kenneth R. Koedinger 2, Alida Skogsholm 2, Adaeze Nwaigwe 2, Robert G.M. Hausmann 1, Anders Weinstein
More informationImproving Action Selection in MDP s via Knowledge Transfer
In Proc. 20th National Conference on Artificial Intelligence (AAAI-05), July 9 13, 2005, Pittsburgh, USA. Improving Action Selection in MDP s via Knowledge Transfer Alexander A. Sherstov and Peter Stone
More informationUniversity of Waterloo School of Accountancy. AFM 102: Introductory Management Accounting. Fall Term 2004: Section 4
University of Waterloo School of Accountancy AFM 102: Introductory Management Accounting Fall Term 2004: Section 4 Instructor: Alan Webb Office: HH 289A / BFG 2120 B (after October 1) Phone: 888-4567 ext.
More informationGeorgetown University at TREC 2017 Dynamic Domain Track
Georgetown University at TREC 2017 Dynamic Domain Track Zhiwen Tang Georgetown University zt79@georgetown.edu Grace Hui Yang Georgetown University huiyang@cs.georgetown.edu Abstract TREC Dynamic Domain
More informationSpeeding Up Reinforcement Learning with Behavior Transfer
Speeding Up Reinforcement Learning with Behavior Transfer Matthew E. Taylor and Peter Stone Department of Computer Sciences The University of Texas at Austin Austin, Texas 78712-1188 {mtaylor, pstone}@cs.utexas.edu
More informationDeveloping True/False Test Sheet Generating System with Diagnosing Basic Cognitive Ability
Developing True/False Test Sheet Generating System with Diagnosing Basic Cognitive Ability Shih-Bin Chen Dept. of Information and Computer Engineering, Chung-Yuan Christian University Chung-Li, Taiwan
More informationReinForest: Multi-Domain Dialogue Management Using Hierarchical Policies and Knowledge Ontology
ReinForest: Multi-Domain Dialogue Management Using Hierarchical Policies and Knowledge Ontology Tiancheng Zhao CMU-LTI-16-006 Language Technologies Institute School of Computer Science Carnegie Mellon
More informationWHY SOLVE PROBLEMS? INTERVIEWING COLLEGE FACULTY ABOUT THE LEARNING AND TEACHING OF PROBLEM SOLVING
From Proceedings of Physics Teacher Education Beyond 2000 International Conference, Barcelona, Spain, August 27 to September 1, 2000 WHY SOLVE PROBLEMS? INTERVIEWING COLLEGE FACULTY ABOUT THE LEARNING
More informationTask Completion Transfer Learning for Reward Inference
Machine Learning for Interactive Systems: Papers from the AAAI-14 Workshop Task Completion Transfer Learning for Reward Inference Layla El Asri 1,2, Romain Laroche 1, Olivier Pietquin 3 1 Orange Labs,
More informationProgress Monitoring for Behavior: Data Collection Methods & Procedures
Progress Monitoring for Behavior: Data Collection Methods & Procedures This event is being funded with State and/or Federal funds and is being provided for employees of school districts, employees of the
More informationGuru: A Computer Tutor that Models Expert Human Tutors
Guru: A Computer Tutor that Models Expert Human Tutors Andrew Olney 1, Sidney D'Mello 2, Natalie Person 3, Whitney Cade 1, Patrick Hays 1, Claire Williams 1, Blair Lehman 1, and Art Graesser 1 1 University
More informationTask Completion Transfer Learning for Reward Inference
Task Completion Transfer Learning for Reward Inference Layla El Asri 1,2, Romain Laroche 1, Olivier Pietquin 3 1 Orange Labs, Issy-les-Moulineaux, France 2 UMI 2958 (CNRS - GeorgiaTech), France 3 University
More informationBEETLE II: a system for tutoring and computational linguistics experimentation
BEETLE II: a system for tutoring and computational linguistics experimentation Myroslava O. Dzikovska and Johanna D. Moore School of Informatics, University of Edinburgh, Edinburgh, United Kingdom {m.dzikovska,j.moore}@ed.ac.uk
More informationDiscriminative Learning of Beam-Search Heuristics for Planning
Discriminative Learning of Beam-Search Heuristics for Planning Yuehua Xu School of EECS Oregon State University Corvallis,OR 97331 xuyu@eecs.oregonstate.edu Alan Fern School of EECS Oregon State University
More informationIntroduction. Chem 110: Chemical Principles 1 Sections 40-52
Introduction Chem 110: Chemical Principles 1 Sections 40-52 Instructor: Dr. Squire J. Booker 302 Chemistry Building 814-865-8793 squire@psu.edu (sjb14@psu.edu) Lectures: Monday (M), Wednesday (W), Friday
More informationAxiom 2013 Team Description Paper
Axiom 2013 Team Description Paper Mohammad Ghazanfari, S Omid Shirkhorshidi, Farbod Samsamipour, Hossein Rahmatizadeh Zagheli, Mohammad Mahdavi, Payam Mohajeri, S Abbas Alamolhoda Robotics Scientific Association
More informationImproving Conceptual Understanding of Physics with Technology
INTRODUCTION Improving Conceptual Understanding of Physics with Technology Heidi Jackman Research Experience for Undergraduates, 1999 Michigan State University Advisors: Edwin Kashy and Michael Thoennessen
More informationLearning Optimal Dialogue Strategies: A Case Study of a Spoken Dialogue Agent for
Learning Optimal Dialogue Strategies: A Case Study of a Spoken Dialogue Agent for Email Marilyn A. Walker Jeanne C. Fromer Shrikanth Narayanan walker@research.att.com jeannie@ai.mit.edu shri@research.att.com
More informationBENCHMARK TREND COMPARISON REPORT:
National Survey of Student Engagement (NSSE) BENCHMARK TREND COMPARISON REPORT: CARNEGIE PEER INSTITUTIONS, 2003-2011 PREPARED BY: ANGEL A. SANCHEZ, DIRECTOR KELLI PAYNE, ADMINISTRATIVE ANALYST/ SPECIALIST
More informationSession 2B From understanding perspectives to informing public policy the potential and challenges for Q findings to inform survey design
Session 2B From understanding perspectives to informing public policy the potential and challenges for Q findings to inform survey design Paper #3 Five Q-to-survey approaches: did they work? Job van Exel
More informationUK Institutional Research Brief: Results of the 2012 National Survey of Student Engagement: A Comparison with Carnegie Peer Institutions
UK Institutional Research Brief: Results of the 2012 National Survey of Student Engagement: A Comparison with Carnegie Peer Institutions November 2012 The National Survey of Student Engagement (NSSE) has
More informationUniversity of Groningen. Systemen, planning, netwerken Bosman, Aart
University of Groningen Systemen, planning, netwerken Bosman, Aart IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document
More informationCase study Norway case 1
Case study Norway case 1 School : B (primary school) Theme: Science microorganisms Dates of lessons: March 26-27 th 2015 Age of students: 10-11 (grade 5) Data sources: Pre- and post-interview with 1 teacher
More informationGCSE Mathematics B (Linear) Mark Scheme for November Component J567/04: Mathematics Paper 4 (Higher) General Certificate of Secondary Education
GCSE Mathematics B (Linear) Component J567/04: Mathematics Paper 4 (Higher) General Certificate of Secondary Education Mark Scheme for November 2014 Oxford Cambridge and RSA Examinations OCR (Oxford Cambridge
More informationNo. 30. CRANFORD, NEW JERSEY, THURSDAY, AUGUST
/. -. THE OCANFORl CTZENAND CHRONCLE. THURSDAY, AUGUST 4,958 S RevewedatLocalMeetng \ Servce has alwas? been the kestone of Rotar,t. Wesle A. Stanger, Sr., told fellow members of the Cranford Rotar Club
More informationEvaluation of a College Freshman Diversity Research Program
Evaluation of a College Freshman Diversity Research Program Sarah Garner University of Washington, Seattle, Washington 98195 Michael J. Tremmel University of Washington, Seattle, Washington 98195 Sarah
More informationNCEO Technical Report 27
Home About Publications Special Topics Presentations State Policies Accommodations Bibliography Teleconferences Tools Related Sites Interpreting Trends in the Performance of Special Education Students
More informationExploration. CS : Deep Reinforcement Learning Sergey Levine
Exploration CS 294-112: Deep Reinforcement Learning Sergey Levine Class Notes 1. Homework 4 due on Wednesday 2. Project proposal feedback sent Today s Lecture 1. What is exploration? Why is it a problem?
More informationOrganising ROSE (The Relevance of Science Education) survey in Finland
25.02.2004 1 Organising ROSE (The Relevance of Science Education) survey in Finland Researchers and support The Survey was organised by the following researchers at the Department of Teacher Education,
More informationISFA2008U_120 A SCHEDULING REINFORCEMENT LEARNING ALGORITHM
Proceedings of 28 ISFA 28 International Symposium on Flexible Automation Atlanta, GA, USA June 23-26, 28 ISFA28U_12 A SCHEDULING REINFORCEMENT LEARNING ALGORITHM Amit Gil, Helman Stern, Yael Edan, and
More informationThe Good Judgment Project: A large scale test of different methods of combining expert predictions
The Good Judgment Project: A large scale test of different methods of combining expert predictions Lyle Ungar, Barb Mellors, Jon Baron, Phil Tetlock, Jaime Ramos, Sam Swift The University of Pennsylvania
More informationA Reinforcement Learning Variant for Control Scheduling
A Reinforcement Learning Variant for Control Scheduling Aloke Guha Honeywell Sensor and System Development Center 3660 Technology Drive Minneapolis MN 55417 Abstract We present an algorithm based on reinforcement
More informationA Study of Metacognitive Awareness of Non-English Majors in L2 Listening
ISSN 1798-4769 Journal of Language Teaching and Research, Vol. 4, No. 3, pp. 504-510, May 2013 Manufactured in Finland. doi:10.4304/jltr.4.3.504-510 A Study of Metacognitive Awareness of Non-English Majors
More informationConnect Microbiology. Training Guide
1 Training Checklist Section 1: Getting Started 3 Section 2: Course and Section Creation 4 Creating a New Course with Sections... 4 Editing Course Details... 9 Editing Section Details... 9 Copying a Section
More informationPhilosophy of Literacy. on a daily basis. My students will be motivated, fluent, and flexible because I will make my reading
Balanced Literacy Summer 2010 Philosophy of Literacy My personal philosophy on literacy in the classroom is to develop good readers who actively interact and make connections with the text in order to
More informationWhen!Identifying!Contributors!is!Costly:!An! Experiment!on!Public!Goods!
!! EVIDENCE-BASED RESEARCH ON CHARITABLE GIVING SPI$FUNDED$ When!Identifying!Contributors!is!Costly:!An! Experiment!on!Public!Goods! Anya!Samek,!Roman!M.!Sheremeta!! University!of!WisconsinFMadison! Case!Western!Reserve!University!&!Chapman!University!!
More information2005 National Survey of Student Engagement: Freshman and Senior Students at. St. Cloud State University. Preliminary Report.
National Survey of Student Engagement: Freshman and Senior Students at St. Cloud State University Preliminary Report (December, ) Institutional Studies and Planning National Survey of Student Engagement
More informationPedagogical Content Knowledge for Teaching Primary Mathematics: A Case Study of Two Teachers
Pedagogical Content Knowledge for Teaching Primary Mathematics: A Case Study of Two Teachers Monica Baker University of Melbourne mbaker@huntingtower.vic.edu.au Helen Chick University of Melbourne h.chick@unimelb.edu.au
More informationschool students to improve communication skills
Motivating middle and high school students to improve communication skills Megan Mahowald, Ph.D. CCC-SLP Indiana University mcmahowa@indiana.edu Case Study High Motivation Low Motivation Behaviors what
More informationDiscourse Processing for Explanatory Essays in Tutorial Applications
Discourse Processing for Explanatory Essays in Tutorial Applications Pamela W. Jordan and Kurt VanLehn Learning Research and Development Center University of Pittsburgh Pittsburgh PA 15260 [pjordan,vanlehn]@pitt.edu
More informationImproved Effects of Word-Retrieval Treatments Subsequent to Addition of the Orthographic Form
Orthographic Form 1 Improved Effects of Word-Retrieval Treatments Subsequent to Addition of the Orthographic Form The development and testing of word-retrieval treatments for aphasia has generally focused
More informationRegret-based Reward Elicitation for Markov Decision Processes
444 REGAN & BOUTILIER UAI 2009 Regret-based Reward Elicitation for Markov Decision Processes Kevin Regan Department of Computer Science University of Toronto Toronto, ON, CANADA kmregan@cs.toronto.edu
More informationPre-Algebra A. Syllabus. Course Overview. Course Goals. General Skills. Credit Value
Syllabus Pre-Algebra A Course Overview Pre-Algebra is a course designed to prepare you for future work in algebra. In Pre-Algebra, you will strengthen your knowledge of numbers as you look to transition
More informationBackwards Numbers: A Study of Place Value. Catherine Perez
Backwards Numbers: A Study of Place Value Catherine Perez Introduction I was reaching for my daily math sheet that my school has elected to use and in big bold letters in a box it said: TO ADD NUMBERS
More informationNATIONAL SURVEY OF STUDENT ENGAGEMENT
NATIONAL SURVEY OF STUDENT ENGAGEMENT 2010 Benchmark Comparisons Report OFFICE OF INSTITUTIONAL RESEARCH & PLANNING To focus discussions about the importance of student engagement and to guide institutional
More informationTHE PENNSYLVANIA STATE UNIVERSITY SCHREYER HONORS COLLEGE DEPARTMENT OF MATHEMATICS ASSESSING THE EFFECTIVENESS OF MULTIPLE CHOICE MATH TESTS
THE PENNSYLVANIA STATE UNIVERSITY SCHREYER HONORS COLLEGE DEPARTMENT OF MATHEMATICS ASSESSING THE EFFECTIVENESS OF MULTIPLE CHOICE MATH TESTS ELIZABETH ANNE SOMERS Spring 2011 A thesis submitted in partial
More informationDoes the Difficulty of an Interruption Affect our Ability to Resume?
Difficulty of Interruptions 1 Does the Difficulty of an Interruption Affect our Ability to Resume? David M. Cades Deborah A. Boehm Davis J. Gregory Trafton Naval Research Laboratory Christopher A. Monk
More informationThe Use of Statistical, Computational and Modelling Tools in Higher Learning Institutions: A Case Study of the University of Dodoma
International Journal of Computer Applications (975 8887) The Use of Statistical, Computational and Modelling Tools in Higher Learning Institutions: A Case Study of the University of Dodoma Gilbert M.
More informationCS 100: Principles of Computing
CS 100: Principles of Computing Kevin Molloy August 29, 2017 1 Basic Course Information 1.1 Prerequisites: None 1.2 General Education Fulfills Mason Core requirement in Information Technology (ALL). 1.3
More informationTD(λ) and Q-Learning Based Ludo Players
TD(λ) and Q-Learning Based Ludo Players Majed Alhajry, Faisal Alvi, Member, IEEE and Moataz Ahmed Abstract Reinforcement learning is a popular machine learning technique whose inherent self-learning ability
More informationSave Children. Can Math Recovery. before They Fail?
Can Math Recovery Save Children before They Fail? numbers just get jumbled up in my head. Renee, a sweet six-year-old with The huge brown eyes, described her frustration this way. Not being able to make
More informationFCE Speaking Part 4 Discussion teacher s notes
Description Brainstorming activity designed to raise students awareness of discussion questions followed by controlled practice and a True/False activity. Time required: Additional materials required:
More informationGrade 6: Module 2A: Unit 2: Lesson 8 Mid-Unit 3 Assessment: Analyzing Structure and Theme in Stanza 4 of If
Grade 6: Module 2A: Unit 2: Lesson 8 Mid-Unit 3 Assessment: Analyzing Structure and This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Exempt third-party
More informationContinual Curiosity-Driven Skill Acquisition from High-Dimensional Video Inputs for Humanoid Robots
Continual Curiosity-Driven Skill Acquisition from High-Dimensional Video Inputs for Humanoid Robots Varun Raj Kompella, Marijn Stollenga, Matthew Luciw, Juergen Schmidhuber The Swiss AI Lab IDSIA, USI
More informationPredicting Students Performance with SimStudent: Learning Cognitive Skills from Observation
School of Computer Science Human-Computer Interaction Institute Carnegie Mellon University Year 2007 Predicting Students Performance with SimStudent: Learning Cognitive Skills from Observation Noboru Matsuda
More informationUsing focal point learning to improve human machine tacit coordination
DOI 10.1007/s10458-010-9126-5 Using focal point learning to improve human machine tacit coordination InonZuckerman SaritKraus Jeffrey S. Rosenschein The Author(s) 2010 Abstract We consider an automated
More informationIMPORTANT STEPS WHEN BUILDING A NEW TEAM
IMPORTANT STEPS WHEN BUILDING A NEW TEAM This article outlines essential steps in forming a new team. These steps are also useful for existing teams that are interested in assessing their format and effectiveness.
More informationBy. Candra Pantura Panlaysia Dr. CH. Evy Tri Widyahening, S.S., M.Hum Slamet Riyadi University Surakarta ABSTRACT
THE EFFECTIVENESS OF MIND MAPPING TECHNIQUE IN TEACHING LEARNING WRITING ON RECOUNT TEXT (An Experimental Study in the Tenth Grade Students of MAN 2 SurakartaIn 2015/2016 Academic Year) By. Candra Pantura
More informationLearning Methods in Multilingual Speech Recognition
Learning Methods in Multilingual Speech Recognition Hui Lin Department of Electrical Engineering University of Washington Seattle, WA 98125 linhui@u.washington.edu Li Deng, Jasha Droppo, Dong Yu, and Alex
More informationFoothill College Fall 2014 Math My Way Math 230/235 MTWThF 10:00-11:50 (click on Math My Way tab) Math My Way Instructors:
This is a team taught directed study course. Foothill College Fall 2014 Math My Way Math 230/235 MTWThF 10:00-11:50 www.psme.foothill.edu (click on Math My Way tab) Math My Way Instructors: Instructor:
More informationEHE. a corn TuesdaY morning 00 the (Xte land about nine miles north and CIle and ooe-quarter west c:i Wayne.
pages sectons NNETY TlRD YEAR 'P N T W Hansen Real Wnner ~ 'ames op... rs h Tuesday Electon Plger Cambrtdp. RaBtm aqd North Platte have been named tbp wnners n the t 968 Com YU81lty mprovement Program.
More informationEvaluation of Hybrid Online Instruction in Sport Management
Evaluation of Hybrid Online Instruction in Sport Management Frank Butts University of West Georgia fbutts@westga.edu Abstract The movement toward hybrid, online courses continues to grow in higher education
More informationCognitive Apprenticeship Statewide Campus System, Michigan State School of Osteopathic Medicine 2011
Statewide Campus System, Michigan State School of Osteopathic Medicine 2011 Gloria Kuhn, DO, PhD Wayne State University, School of Medicine The is a method of teaching aimed primarily at teaching the thought
More informationStephanie Ann Siler. PERSONAL INFORMATION Senior Research Scientist; Department of Psychology, Carnegie Mellon University
Stephanie Ann Siler PERSONAL INFORMATION Senior Research Scientist; Department of Psychology, Carnegie Mellon University siler@andrew.cmu.edu Home Address Office Address 26 Cedricton Street 354 G Baker
More informationChinese Language Parsing with Maximum-Entropy-Inspired Parser
Chinese Language Parsing with Maximum-Entropy-Inspired Parser Heng Lian Brown University Abstract The Chinese language has many special characteristics that make parsing difficult. The performance of state-of-the-art
More informationLevel 1 Mathematics and Statistics, 2015
91037 910370 1SUPERVISOR S Level 1 Mathematics and Statistics, 2015 91037 Demonstrate understanding of chance and data 9.30 a.m. Monday 9 November 2015 Credits: Four Achievement Achievement with Merit
More informationAssociation Between Categorical Variables
Student Outcomes Students use row relative frequencies or column relative frequencies to informally determine whether there is an association between two categorical variables. Lesson Notes In this lesson,
More informationCOMPUTATIONAL COMPLEXITY OF LEFT-ASSOCIATIVE GRAMMAR
COMPUTATIONAL COMPLEXITY OF LEFT-ASSOCIATIVE GRAMMAR ROLAND HAUSSER Institut für Deutsche Philologie Ludwig-Maximilians Universität München München, West Germany 1. CHOICE OF A PRIMITIVE OPERATION The
More information6 Financial Aid Information
6 This chapter includes information regarding the Financial Aid area of the CA program, including: Accessing Student-Athlete Information regarding the Financial Aid screen (e.g., adding financial aid information,
More informationThe Moodle and joule 2 Teacher Toolkit
The Moodle and joule 2 Teacher Toolkit Moodlerooms Learning Solutions The design and development of Moodle and joule continues to be guided by social constructionist pedagogy. This refers to the idea that
More informationConcept Acquisition Without Representation William Dylan Sabo
Concept Acquisition Without Representation William Dylan Sabo Abstract: Contemporary debates in concept acquisition presuppose that cognizers can only acquire concepts on the basis of concepts they already
More informationKLI: Infer KCs from repeated assessment events. Do you know what you know? Ken Koedinger HCI & Psychology CMU Director of LearnLab
KLI: Infer KCs from repeated assessment events Ken Koedinger HCI & Psychology CMU Director of LearnLab Instructional events Explanation, practice, text, rule, example, teacher-student discussion Learning
More informationHow to set up gradebook categories in Moodle 2.
How to set up gradebook categories in Moodle 2. It is possible to set up the gradebook to show divisions in time such as semesters and quarters by using categories. For example, Semester 1 = main category
More information