130 research outputs found
Optimal information exchange policies in integrated product development
No full textThis article considers information exchange in an Integrated Product Development (IPD) environment. First, a dynamic programming model is formulated that is able to capture upstream partial information flow in a two-activity IPD process. A simple threshold policy is derived that aids the downstream activity in deciding whether to consider or ignore this upstream information as a function of information quality and its associated setup and rework penalties. Then, this formulation is expanded to model analytically, for the first time, information flow in a three-activity IPD process. In this case, the focus is on aiding the midstream activity in deciding whether to consider or ignore partial upstream information, taking into consideration downstream concerns. Because it is difficult to derive threshold policies in this case, the dynamic program has to be solved directly and then an extensive Monte Carlo simulation study is performed to analyze the behavior of the optimal policy. The simulation results suggest several important insights regarding the timing and frequency of considering partial information in an IPD environment. © 2013 Taylor and Francis Group, LLC.Banker RD, 2006, INFORM SYST RES, V17, P352, DOI 10.1287-isre.1060.0104; Bardhan IR, 2007, INFORM TECHNOL MANAG, V8, P167, DOI 10.1007-s10799-007-0013-y; Bazaraa M., 1993, NONLINEAR PROGRAMMIN; Clark K. T., 1991, PRODUCT DEV PERFORMA; EISENHARDT KM, 1995, ADMIN SCI QUART, V40, P84, DOI 10.2307-2393701; Gerwin D, 2002, MANAGE SCI, V48, P938, DOI 10.1287-mnsc.48.7.938.2818; HA AY, 1995, MANAGE SCI, V41, P1431, DOI 10.1287-mnsc.41.9.1431; Hauptman O, 1999, RandD MANAGE, V29, P179, DOI 10.1111-1467-9310.00128; Hoegl M, 2004, ORGAN SCI, V15, P38, DOI 10.1287-orsc.1030.0053; Joglekar NR, 2001, MANAGE SCI, V47, P1605, DOI 10.1287-mnsc.47.12.1605.10240; Krishnan V, 2001, MANAGE SCI, V47, P1, DOI 10.1287-mnsc.47.1.1.10668; Krishnan V, 1997, MANAGE SCI, V43, P437, DOI 10.1287-mnsc.43.4.437; Lin J, 2010, EUR J OPER RES, V201, P737, DOI 10.1016-j.ejor.2009.03.040; Liu DT, 2001, COMPUT IND, V44, P251, DOI 10.1016-S0166-3615(01)00072-0; Loch CH, 1998, MANAGE SCI, V44, P1032, DOI 10.1287-mnsc.44.8.1032; Luenberger D. G., 1998, INVESTMENT SCI; McDonough EF, 2000, J PROD INNOVAT MANAG, V17, P221, DOI 10.1016-S0737-6782(00)00041-2; Nambisan S, 2003, MIS QUART, V27, P1; Nambisan S, 2009, ANN INFORM SYST, V5, P1, DOI 10.1007-978-1-4419-1081-3_1; Roemer TA, 2000, OPER RES, V48, P858, DOI 10.1287-opre.48.6.858.12396; Terwiesch C, 1999, MANAGE SCI, V45, P455, DOI 10.1287-mnsc.45.4.455; Yassine A, 2004, PROD PLAN CONTROL, V15, P422, DOI [10.1080-0953728042000238782, 10.1080-095372804200238782]; Yassine A, 2003, RES ENG DES, V14, P145, DOI 10.1007-S00163-003-0036-2; Yassine AA, 2008, EUR J OPER RES, V184, P311, DOI 10.1016-j.ejor.2006.10.0420
Reconocimiento de entidades nombradas en textos árabes
No full textTesis doctoral en Informática realizada por Yassine Benajiba y dirigida por el doctor Paolo Rosso (Univ. Politécnica de Valencia). El acto de defensa de tesis tuvo lugar en Valencia en Mayo de 2009 ante el tribunal formado por los doctores Felisa Verdejo (UNED), Mona Diab (Columbia Univ.), Imed Zitouni (IBM T.J. Watson Research Center), Horacio Rodríguez (Univ. Politécnica de Cataluña) y Encarna Segarra (Univ. Politécnica de Valencia). La calificación obtenida fue Sobresaliente Cum Laude.PhD thesis in Computer Science written by Yassine Benajiba under the supervision of Dr Paolo Rosso (Univ. Politécnica de Valencia). The author was examined in May 2009 in Valencia by the committee formed by Felisa Verdejo (UNED), Mona Diab (Columbia Univ.), Imed Zitouni (IBM T.J. Watson Research Center), Horacio Rodríguez (Univ. Politécnica de Cataluña) and Encarna Segarra (Univ. Politécnica de Valencia). The grade obtained was Cum Laude.This PhD thesis was supported by an AECI scholarship
Integrating Universities' Thesis and Research Deposit Mandates
No full textA growing number of universities are beginning to require the digital deposit of their thesis and dissertation output in their institutional repositories. At the same time, a growing number of universities as well as research funders are beginning to mandate that all refereed research must be deposited too. This makes for a timely synergy between the practices of the younger and older generation of researchers as the Open Access era unfolds. It also maximizes the uptake, usage and impact of university research input at all stages, as well as providing rich and powerful new metrics to monitor and reward research productivity and impact. It is important to integrate universities' ETD and research output repositories, mandates and metrics as well as to provide the mechanism for those deposits that may need to be made Closed Access rather than Open Access: Repositories need to implement the "email eprint request" Button for all Closed Access Deposits. Any would-be user webwide, having reached the metadata of a Closed Access Deposit can, with one click, request an eprint for research purposes; the author instantly receives an automatic email and can then, again with one click, authorize the automatic emailing of one copy to the user by the repository software. This feature is important for fulfilling immediate research usage needs during any journal-article embargo period, and it also gives the authors of dissertations they hope to publish as books a way to control who has access to the dissertation. Digital dissertations will also benefit from the reference-linking and book-citation metrics that will be provided by harvesters of the distributed institutional repository metadata (which will also include the metadata and reference lists of all university book output). Dissertation downloads as well as eprint-requests will also provide useful new research impact metrics
A knowledge-driven, network-based computational framework for product development systems
No full textToday's fast-paced product development (PD) environment brings many new challenges to the PD community. These challenges are mainly due to a drastic increase in the scale and complexity of engineered systems, which require the collaboration of functionally and geographically distributed resources within and outside a firm's boundary. To address these new challenges, this paper proposes a novel theoretical and computational framework for an enterprise-wide PD management system. The proposed framework considers an integrative view of the various dependencies that co-exist in three PD domains (i.e., people, products, and processes). Additionally, it provides a computational tool that links them together in a succinct and tractable way and provides an analysis method for assessing their influence on shaping the product development process. Using this framework, we suggest that the characteristics of how an organization acquire data, interpret information, and apply knowledge will impact the final architecture of a product. We demonstrate this framework by analyzing the development efforts for a software project called ROBOCODE. Copyright © 2013 by ASME.Abdel-Hamid T., 1991, SOFTWARE PROJECT DYN; Ackoff R.L., 1989, J APPL SYSTEMS ANAL, V16, P3, DOI DOI 10.1002-9781444303179.CH3; Aguilar F, 1967, SCANNING BUSINESS EN; Alexander C., 1964, NOTES SYNTHESIS FORM; Allen T., 1997, 3983 MIT SLOAN SCH M; Baldwin C. Y., 2000, DESIGN RULES, V1; Barkmeyer E.J, 1997, 5939 NISTIR; Bedward D., 2004, MANAGING INFORM CORE; BELKIN NJ, 1976, J AM SOC INFORM SCI, V27, P197, DOI 10.1002-asi.4630270402; Berry MW, 2005, SOFTW ENVIRON TOOLS, V17, P1, DOI 10.1137-1.9780898718164; Borgatti S.P., 2002, UCINET WINDOWS SOFTW; Borgatti SP, 2003, MANAGE SCI, V49, P432, DOI 10.1287-mnsc.49.4.432.14428; Bradley J., 2009, THESIS U ILLINOIS UR; Braha D., 2001, DATA MINING DESIGN M; Browning TR, 2002, IEEE T ENG MANAGE, V49, P443, DOI 10.1109-TEM.2002.806710; BUCKLAND MK, 1991, J AM SOC INFORM SCI, V42, P351, DOI 10.1002-(SICI)1097-4571(199106)42:5351::AID-ASI53.0.CO;2-3; Collier W., 1997, INTEGRAL PIM STRATEG; CONWAY ME, 1968, DATAMATION, V14, P28; Danilovic M., 2007, International Journal of Project Management, V25, DOI 10.1016-j.ijproman.2006.11.003; Davis G. B., 1985, MANAGEMENT INFORM SY; Debons A., 1988, INFORM SCI INTEGRATE; Eppinger S., 2001, P INT C ENG DES ICED; Ess Charles, 2004, COMPANION DIGITAL HU, P132, DOI 10.1002-9780470999875.ch12; Ford DN, 2003, CONCURRENT ENG-RES A, V11, P177, DOI 10.1177-106329303038031; Gokpinar B, 2010, MANAGE SCI, V56, P468, DOI 10.1287-mnsc.1090.1117; Gruninger M, 2003, AI MAG, V24, P63; Halladay S. M., 2006, IEEE P 39 HAW INT C; Hellgren B., 1995, SCAND J MGT, V11, P377, DOI 10.1016-0956-5221(95)00020-V; Hoetker G, 2006, STRATEGIC MANAGE J, V27, P501, DOI 10.1002-smj.528; HOLLAND PW, 1972, AM J SOCIOL, V77, P1205, DOI 10.1086-225266; King N, 1996, IEEE T ENG MANAGE, V43, P189, DOI 10.1109-17.509984; Kogut B, 2001, OXFORD REV ECON POL, V17, P248, DOI 10.1093-oxrep-17.2.248; Krackhardt D., 1998, P 1998 INT S COMM CO, P113; Levitt RE, 1999, MANAGE SCI, V45, P1479, DOI 10.1287-mnsc.45.11.1479; Lindemann U, 2007, Future of Product Development, P351, DOI 10.1007-978-3-540-69820-3_35; MORELLI MD, 1995, IEEE T ENG MANAGE, V42, P215, DOI 10.1109-17.403739; PAPA MJ, 1990, COMMUN RES, V17, P344, DOI 10.1177-009365090017003004; Parashar S., 2005, 46 AIAA ASME ASCE AH; Poli R., 2001, THESIS U UTRECHT NET; Rechtin E., 1997, ART SYSTEMS ARCHITEC; Sako M., 2003, BUSINESS SYSTEM INTE; Salamon W., 1994, QUALITY CHARACTERIST; Sanchez R., 1996, IEEE T ENG MANAGE, V25, P50; Sangal N., 2005, 20 ACM SIGPLAN C OBJ; Sharman D., 2002, ASME 2002 INT DES EN; Shooter SB, 2000, ENG COMPUT-GERMANY, V16, P178, DOI 10.1007-s003660070004; Simon H. A., 1969, SCI ARTIFICIAL; Sosa ME, 2004, MANAGE SCI, V50, P1674, DOI 10.1287-mnsc.1040.0289; Sosa ME, 2008, RES ENG DES, V19, P47, DOI 10.1007-s00163-007-0039-5; Stonier T., 1997, INFORM MEANING EVOLU; Tardy C., 1988, HDB STUDY HUMAN COMM, P107; Wasserman S., 1994, SOCIAL NETWORK ANAL; Wheelwright S. C., 1992, REVOLUTIONIZING PROD; Whitney D., 2004, ESDWP200407; Wu T, 2004, J COMPUT INF SCI ENG, V4, P281, DOI 10.1155-1.1814385; Yassine A, 2004, PROD PLAN CONTROL, V15, P422, DOI [10.1080-0953728042000238782, 10.1080-095372804200238782]; Yassine A, 2003, J ENG DESIGN, V14, P377, DOI 10.1080-0954482031000091103; Yassine A, 2003, RES ENG DES, V14, P145, DOI 10.1007-S00163-003-0036-2; Yu TL, 2007, RES ENG DES, V18, P91, DOI 10.1007-s00163-007-0030-1; Zins C, 2007, J AM SOC INF SCI TEC, V58, P479, DOI 10.1002-asi.205080
Parametric design adaptation for competitive products
No full textVery often product development is seen as a process where designers iterate through several design cycles until they converge upon a design that satisfies all of the necessary requirements-design within a single generation. If one takes the view that products change (i.e. adapt and evolve), a broader view must be adopted to capture the drivers of design adaptation across multiple product generations. This paper offers a new multi-generation conceptual framework of parametric design adaptation for consumer products, called the Artisan-Patron (AP) framework, and a complementary computational model. The AP framework captures the interaction between manufacturers (the Artisan) and consumers (the Patron) by structuring the various relevant information (e.g., consumer taste, government policy, cost of raw materials, etc.). Additionally, based on this framework, a corresponding computational model is developed, which allows engineers to find optimal settings for the design variables in a dynamic multi-generation environment. The utility of the conceptual framework and the computational model is demonstrated by considering the parametric design adaptation of the automobile with respect to two design parameters- engine horsepower and weight-based on historical automotive industry data. © Springer Science+Business Media, LLC 2010.Ahmad S., 2005, TRANSPORT RES REC, P1, DOI 10.3141-1941-01; Angel A., 2008, SYSTEMS ENG J, V9, P125; Asiedu P., 1998, INT J PROD RES, V36, P883; Ben-Akiva M., 1985, DISCRETE CHOICE ANAL; Ben-Arieh D, 2003, INT J PROD ECON, V83, P169, DOI 10.1016-S0925-5273(02)00323-7; BERRYMAN AA, 1992, ECOLOGY, V73, P1530, DOI 10.2307-1940005; Cardoso MF, 1997, COMPUT CHEM ENG, V21, P1349, DOI 10.1016-S0098-1354(97)00015-X; Castagne S, 2008, RES ENG DES, V18, P149, DOI 10.1007-s00163-007-0042-x; CBO-Congressional Budget Office, 2002, RED GAS CONS 3 POL O; CLARK KB, 1985, RES POLICY, V14, P235, DOI 10.1016-0048-7333(85)90007-1; Cook H. E., 2005, DESIGN 6 SIGMA STRAT; Cook H. E., 1997, PRODUCT MANAGEMENT V; Cook HE, 2001, RES ENG DES, V13, P42, DOI 10.1007-s001630100004; Cuenca R., 2000, COMP INDIRECT COST M; Donndelinger J., 1997, SAE T J PASSENGER CA, V106, P1263; Eckert C, 2004, RES ENG DES, V15, P1, DOI 10.1007-s00163-003-0031-7; Eppinger SD, 1997, RES ENG DES, V9, P112, DOI 10.1007-BF01596486; FARS-Fatality Analysis Reporting System, 2008, FARS FATALITY ANAL R; Ferguson S., 2007, ASME DES AUT C SEPT; Ferreira EC, 2001, COMPUT APPL ENG EDUC, V9, P49, DOI 10.1002-cae.1005; Fricke E., 2005, Systems Engineering, V8, DOI 10.1002-sys.20039; Georgiopoulos P, 2005, J MECH DESIGN, V127, P358, DOI 10.1115-1.1862679; Hazelrigg GA, 2003, ENG OPTIMIZ, V35, P103, DOI 10.1080-0305215031000097059; Hellman K. H., 2004, LIGHT DUTY AUTOMOTIV; Hickernell F., 1997, OR T, V1, P1; IIHS-Insurance Institute for Highway Safety, 1998, NEW STUD REL VEH WEI; Jedidi K, 2002, MANAGE SCI, V48, P1350, DOI 10.1287-mnsc.48.10.1350.272; Jensen P. A., 2003, OPERATIONS RES MODEL; Kirchain R., 2001, ENCY MAT SCI ENG, P1718; KOHLI R, 1991, J MARKETING RES, V28, P347, DOI 10.2307-3172870; Krishnan V, 1995, J MECH DESIGN, V117, P491, DOI 10.1115-1.2826709; Liu HF, 2009, J INTELL MANUF, V20, P401, DOI 10.1007-s10845-008-0114-4; Luce R. D., 1959, INDIVIDUAL CHOICE BE; Markel T, 2002, J POWER SOURCES, V110, P255, DOI 10.1016-S0378-7753(02)00189-1; Marti R, 2003, HDB METAHEURISTICS, P355; Marti R., 2009, HDB METAHEURISTICS; Massey G. R., 1999, J PRODUCT BRAND MANA, V8, P301, DOI 10.1108-10610429910284292; Mayr E., 2001, WHAT EVOLUTION IS; McConville G. P., 1996, 960004 SAE; Michalek JJ, 2004, J MECH DESIGN, V126, P1062, DOI 10.1115-1.1804195; MILLER GA, 1956, PSYCHOL REV, V63, P81, DOI 10.1037--0033-295X.101.2.343; Mills V, 2002, VEHICLE SYST DYN, V37, P283, DOI 10.1076-vesd.37.4.283.3530; Monroe E. M., 1997, 97076 SAE; Montgomery D. C., 2000, DESIGN ANAL EXPT; Nelson R., 1982, EVOLUTIONARY THEORY; Olewnik A, 2006, J ENG DESIGN, V17, P75, DOI 10.1080-09544820500274019; OTA-Office of Technology Assessment, 1995, OTAETI638; Otto KN, 1998, RES ENG DES, V10, P226, DOI 10.1007-s001639870003; Papalambros PY, 2002, COMPUT AIDED DESIGN, V34, P939, DOI 10.1016-S0010-4485(01)00148-8; Pepall L., 2002, IND ORG CONT THEORY; Petroski Henry, 2003, SMALL THINGS CONSIDE; Pontikakis G, 2001, P I MECH ENG D-J AUT, V215, P1005, DOI 10.1243-0954407011528572; Shooter SB, 2000, ENG COMPUT-GERMANY, V16, P178, DOI 10.1007-s003660070004; Silver MR, 2007, SYST ENG, V10, P167, DOI 10.1002-sys.20069; Simon H. A., 1996, SCI ARTIFICIAL; Suh ES, 2007, RES ENG DES, V18, P67, DOI 10.1007-s00163-007-0032-z; Suh NP, 1999, RES ENG DES, V11, P116, DOI 10.1007-PL00003883; TELLIS GJ, 1981, J MARKETING, V45, P125, DOI 10.2307-1251480; Thomke SH, 1998, MANAGE SCI, V44, P743, DOI 10.1287-mnsc.44.6.743; Ullman D. G., 1993, DESIGN THEORY METHOD, V53, P91; Utterback JM., 1996, MASTERING DYNAMICS I; Ward's Dealer Business, 2004, WARDS DEALER BU 0901, P9; Wassenaar HJ, 2003, J MECH DESIGN, V125, P490, DOI 10.1115-1.1587156; Wheelwright S.C., 1994, INT MARKET REV, V11, P32, DOI 10.1108-02651339410057509; Whitney D. E., 1990, Research in Engineering Design, V2, DOI 10.1007-BF02029818; Wilke W. L., 1994, CONSUMER BEHAV; Wissmann L., 2005, ASME DES THEOR METH; Wright I., 1997, DESIGN STUDIES, V18, P33, DOI 10.1016-S0142-694X(96)00029-4; Yassine A, 2003, RES ENG DES, V14, P145, DOI 10.1007-S00163-003-0036-2; Zacharias NA, 2008, J INTELL MANUF, V19, P131, DOI 10.1007-s10845-007-0069-x10
Factors for Success in SMEs: A Perspective from Tangier
No full textPurpose
It is becoming increasingly difficult to ignore the importance of the small and medium enterprise (SME) sector in the economic and social development of a country. In view of their identified importance, this present study aims to develop a clearer understanding of the factors that influence success of small and medium enterprises in Tangier, as perceived by local owner-managers.
Design
The initial research framework, which emanated from a systematic literature review, was empirically investigated using a two-stage design, which incorporated both quantitative and qualitative approaches. The rationale behind the two-stage methodology was firstly to avoid common-method bias; and, secondly, to seek to illuminate findings arising from the survey by providing individual insights. Approaches were employed in succession with the findings from the quantitative phase informing the qualitative phase. Initially, a paper and online survey questionnaire was administered to a population of 365 industrial SMEs, identified from the official website of the Ministry of Commerce, Industry and New Technologies, the 2010 directory of the Association of the Industrial Zone of Tangier as well as its official website, and the website of the Tangier free zone. This survey was used in order to validate the initial conceptual framework and gain some insights on the perceptions of owner-managers of the factors influencing the performance of SMEs. Following the quantitative phase, fifteen in-depth face-to-face semi-structured interviews were conducted with selected owner-managers of SMEs, forming a judgmental sample, to explore their experiences, beliefs, and attitudes with respect to the drivers of success.
Findings & conclusions
The study found that there are three generalized influences on the success of SMEs based on the Tangier entrepreneurs' perceptions: the 'owner-manager attributes' with an emphasis on the language skills factor; partnership working with an emphasis on the financial and networking partnership; and business characteristics with a focus on the location factor, which was mainly associated with the ‘free zones’. The existence of these ‘free zones’ was a key factor in selecting Tangier as a location for this study
Une approche modulaire pour l’apprentissage par imitation générique à l’aide d’une représentation spatio-temporelle des démonstrations basée sur les graphes : application à l’apprentissage robotique
No full textReinforcement Learning and Imitation Learning allow robots to learn how to perform tasks independently, without the need for explicit instructions. This thesis examines both methods and integrates them into a modular and generic framework for solving the imitation learning from observation problem. The approach is implemented in two stages, beginning with learning a representation model that captures the spatial and temporal features of observed demonstrations, followed by applying an off-the-shelf RL algorithm with a taskagnostic reward function to learn the imitation policy. Experimental results indicate that the proposed method outperforms state-of-the-art methods and exhibits promising generalisation capabilities across a range of manipulation tasks, surpassing generative methods in most instances.L’apprentissage par renforcement et l’apprentissage par imitation permettent aux robots d’apprendre à effectuer des tâches de manière autonome, sans avoir besoin d’instructions explicites. Cette thèse examine les deux méthodes et les intègre dans un cadre modulaire et générique pour résoudre le problème d’apprentissage par imitation à partir d’observations. L’approche est mise en œuvre en deux étapes, en commençant par apprendre un modèle de représentation qui capture les caractéristiques spatiales et temporelles des démonstrations observées, suivi de l’application d’un algorithme RL prêt à l’emploi avec une fonction de récompense générique pour apprendre la politique d’imitation. Les résultats expérimentaux indiquent que la méthode proposée surpasse les méthodes de pointe et présente des capacités de généralisation prometteuses pour une gamme de tâches de manipulation, dépassant les méthodes génératives dans la plupart des cas
Une approche modulaire pour l’apprentissage par imitation générique à l’aide d’une représentation spatio-temporelle des démonstrations basée sur les graphes : application à l’apprentissage robotique
No full textReinforcement Learning and Imitation Learning allow robots to learn how to perform tasks independently, without the need for explicit instructions. This thesis examines both methods and integrates them into a modular and generic framework for solving the imitation learning from observation problem. The approach is implemented in two stages, beginning with learning a representation model that captures the spatial and temporal features of observed demonstrations, followed by applying an off-the-shelf RL algorithm with a taskagnostic reward function to learn the imitation policy. Experimental results indicate that the proposed method outperforms state-of-the-art methods and exhibits promising generalisation capabilities across a range of manipulation tasks, surpassing generative methods in most instances.L’apprentissage par renforcement et l’apprentissage par imitation permettent aux robots d’apprendre à effectuer des tâches de manière autonome, sans avoir besoin d’instructions explicites. Cette thèse examine les deux méthodes et les intègre dans un cadre modulaire et générique pour résoudre le problème d’apprentissage par imitation à partir d’observations. L’approche est mise en œuvre en deux étapes, en commençant par apprendre un modèle de représentation qui capture les caractéristiques spatiales et temporelles des démonstrations observées, suivi de l’application d’un algorithme RL prêt à l’emploi avec une fonction de récompense générique pour apprendre la politique d’imitation. Les résultats expérimentaux indiquent que la méthode proposée surpasse les méthodes de pointe et présente des capacités de généralisation prometteuses pour une gamme de tâches de manipulation, dépassant les méthodes génératives dans la plupart des cas
Corrigendum: Performance evaluation of five commercial assays in assessing seroprevalence of HEV antibodies among blood donors
No full textThe affiliation for author Lukman Thalib was incorrectly listed as number 5. It should have been listed as number 3. Please see corrected author and affiliation list below:
Enas S. Al-Absi,1,2 Duaa W. Al- Sadeq,1 Manaf H. Younis,3 Hadi M. Yassine,2 Omnya M. Abdalla,1 Areej G. Mesleh,1 Tameem A. Hadwan,1 Joshua O. Amimo,4,5 Lukman Thalib,3 and Gheyath K. Nasrallah1,2,*
1 Department of Biomedical Science, College of Health Sciences, Qatar University, Doha, Qatar
2 Biomedical Research Center, Qatar University, Doha, Qatar
3 Department of Public Health, College of Health Sciences, Qatar University, Doha, Qatar
4 Department of Animal Production, Faculty of Veterinary Medicine University of Nairobi, Nairobi, Kenya
5 Biosciences of Eastern and Central Africa-International Livestock Research Institute (BecA-ILRI), Nairobi, Kenya
The authors apologize for any inconvenience caused.
© 2019 The AuthorsNo Full Tex
Une approche modulaire pour l’apprentissage par imitation générique à l’aide d’une représentation spatio-temporelle des démonstrations basée sur les graphes : application à l’apprentissage robotique
No full textL’apprentissage par renforcement et l’apprentissage par imitation permettent aux robots d’apprendre à effectuer des tâches de manière autonome, sans avoir besoin d’instructions explicites. Cette thèse examine les deux méthodes et les intègre dans un cadre modulaire et générique pour résoudre le problème d’apprentissage par imitation à partir d’observations. L’approche est mise en œuvre en deux étapes, en commençant par apprendre un modèle de représentation qui capture les caractéristiques spatiales et temporelles des démonstrations observées, suivi de l’application d’un algorithme RL prêt à l’emploi avec une fonction de récompense générique pour apprendre la politique d’imitation. Les résultats expérimentaux indiquent que la méthode proposée surpasse les méthodes de pointe et présente des capacités de généralisation prometteuses pour une gamme de tâches de manipulation, dépassant les méthodes génératives dans la plupart des cas.Reinforcement Learning and Imitation Learning allow robots to learn how to perform tasks independently, without the need for explicit instructions. This thesis examines both methods and integrates them into a modular and generic framework for solving the imitation learning from observation problem. The approach is implemented in two stages, beginning with learning a representation model that captures the spatial and temporal features of observed demonstrations, followed by applying an off-the-shelf RL algorithm with a taskagnostic reward function to learn the imitation policy. Experimental results indicate that the proposed method outperforms state-of-the-art methods and exhibits promising generalisation capabilities across a range of manipulation tasks, surpassing generative methods in most instances
- …
