Apprentissage des réseaux de neurones profonds et applications en traitement automatique de la langue naturelle

En apprentissage automatique, domaine qui consiste à utiliser des données pour apprendre une solution aux problèmes que nous voulons confier à la machine, le modèle des Réseaux de Neurones Artificiels (ANN) est un outil précieux. Il a été inventé voilà maintenant près de soixante ans, et pourtant, il est encore de nos jours le sujet d'une recherche active. Récemment, avec l'apprentissage profond, il a en effet permis d'améliorer l'état de l'art dans de nombreux champs d'applications comme la vision par ordinateur, le traitement de la parole et le traitement des langues naturelles. La quantité toujours grandissante de données disponibles et les améliorations du matériel informatique ont permis de faciliter l'apprentissage de modèles à haute capacité comme les ANNs profonds. Cependant, des difficultés inhérentes à l'entraînement de tels modèles, comme les minima locaux, ont encore un impact important. L'apprentissage profond vise donc à trouver des solutions, en régularisant ou en facilitant l'optimisation. Le pré-entraînnement non-supervisé, ou la technique du ``Dropout'', en sont des exemples. Les deux premiers travaux présentés dans cette thèse suivent cette ligne de recherche. Le premier étudie les problèmes de gradients diminuants/explosants dans les architectures profondes. Il montre que des choix simples, comme la fonction d'activation ou l'initialisation des poids du réseaux, ont une grande influence. Nous proposons l'initialisation normalisée pour faciliter l'apprentissage. Le second se focalise sur le choix de la fonction d'activation et présente le rectifieur, ou unité rectificatrice linéaire. Cette étude a été la première à mettre l'accent sur les fonctions d'activations linéaires par morceaux pour les réseaux de neurones profonds en apprentissage supervisé. Aujourd'hui, ce type de fonction d'activation est une composante essentielle des réseaux de neurones profonds. Les deux derniers travaux présentés se concentrent sur les applications des ANNs en traitement des langues naturelles. Le premier aborde le sujet de l'adaptation de domaine pour l'analyse de sentiment, en utilisant des Auto-Encodeurs Débruitants. Celui-ci est encore l'état de l'art de nos jours. Le second traite de l'apprentissage de données multi-relationnelles avec un modèle à base d'énergie, pouvant être utilisé pour la tâche de désambiguation de sens.

Farm Programmes of Electronic Media: A Comparative Study of Audience Perception in Kerala

Farm communication and extension programs are vital part of the farm development attempts. Electronic media plays a major role in farm extension activities. Kerala, the consumer state, which was a complete agricultural state in pre-independence period, is the sprouting land of agricultural extension and publication activities in print media. Later AIR (All India Radio) farm programs and farm broadcasting of Doordarshan enriched the role of electronic media in farm extension activities. The media saturated southern state of India received this new electronic media farm communication revolution whole heartedly. However, after 1990, Kerala witnessed a flood of private T V channels and currently there are 24 channels in this regional language, named Malayalam. All major news and entertainment channels are broadcasting farm programs. Farm programs of AIR and Doordarshan, broadcasted in Malayalam language, have been well accepted to the farmers‘ in Kerala. However, post-independence period, witnessed the formation of Kerala state in Indian Union and the first ballot-elected communist Government started its administration. After the land reform bills, the state witnessed a gradual decrease in agricultural production. Even if it is not reflected much in the attitude and practices of farm community and farm broadcast of traditional electronic broadcasting, a change is observable after the post-liberalization era of India. Private Television channels, which were focused on entertainment value of programs, started broadcasting farm programs and the parameters of program production went through certain changes. In this situation, there is ample relevance for a study about the farm programs of electronic media in terms of a comparative study of audience perception. The study is limited in the state of Kerala as it is the most media saturated state in India. The study analyzes the rate, nature and scope of adoption of farming methods transmitted through electronic media (T.V. and Radio) in Malayalam language.All kinds of Farm programs including comprehensive program serials, success stories, seasonal cropping methods, experts opinion, been analyzed on the basis of the following objectives.  To find whether propagating new farm methods through farm programs in electronic media or the availability of adequate infrastructure and economic factors make a farmer to adopt a new farming method.  To find which electronic media has more influence on farmers to adopt agricultural programs.  To find which form of electronic media gets better feedback from farmers  To find out whether the programs of T.V. or Radio is more acceptable to farmers than the print media.  To find whether farmers gets the message through their preferred medium for the message. The researcher recorded opinions from a panel of agricultural officers, farm Information officers, agro extension researchers and experts. According to their opinions and guidelines, a pilot study is designed and conducted in Kanjikuzhy Panchayath, in Alappuzha district, Kerala. The Panchayath is selected by considering its ideal nature of being the sample for a social Science research. Besides, the nature of farming in the Panchayath, which devoid of the cultivation of cash crops also supported its sample value. As per the observations from the pilot study, researcher confirmed the Triangulation method as the methodology of research. The questionnaire survey, being the primary part contained 42 Questions with 6 independent and 32 dependent variables. The survey is conducted among 400 respondents in Idukki, Alappuzha and Pathanamthitta districts considering geographical differences and distribution of different types of crops. The response from a total of 360 respondents, 120 from each district, finally selected for tabulation and data analysis.The data analysis, based on percentage analysis, along with the results from focus group discussion among a selected group of 20 farmers, together produced the results as follows. Farmers, who are the audience of farm programs, have a very serious approach towards the medium. They are maintaining a critical point of view towards the content of the programs. Farmers are reasonably aware about the financial side of the programs and the monitory aspirations of both private and Government owned Television channels. Even though, the farmers are not aware on the technical terminology and jargons, they have ideas about success stories, program serials and they are even informed about channels are not maintaining an audience research section like AIR. Though the farmers accept Doordarshan as the credential source of farm information and methods, they are inclined to the entertainment value of programs too. They prefer to have more entertainment value for the programs of Doordarshan. Surprisingly, they have very solid suggestions on even about the shots which add entertainment value to the farm broadcasting methods of Doordarshan. Farmers are very much aware about the fact that media is just an instrument for inspiration and persuasion. They strongly believe that the source of information and new methods is agricultural research and an effective change happens only when there are adequate infrastructure and marketing facilities, along with the proper support from Government agricultural guideline and support systems like Krishi Bhavans. They strongly believe that media alone cannot create any magic in increasing agricultural production. Farmers are pointing out the lack of response to the feedback and queries of farmers on farming methods, as an evidence for the difference in levels of commitment of Government and private owned Television channels.Farmers are still perceiving AIR farm programs are far more committed to farmers and farming than any other electronic medium. However, they are seriously lacking Radio receivers with medium wave reception facility. Farmers perceive that the farming methods on new crops are more adoptable than the farming methods of traditional crops in both private and Government owned Television channels. There are multiple factors behind this observation from farmers. Farmers changed in terms of viewing habits and they prefer success stories, which are totally irrelevant and they even think that such stories encourage people to go for farming and they opined that such stories are good sources of inspiration. However, they are all very much sure about the importance and particular about the presence of entertainment factor even in farm programs. Farmers expect direct interaction of any expert of the new farming method to implement the method in their agriculture practices. Though introduction of a new idea in the T.V. is acceptable, farmers need the direct instruction of expert on field to start implementing the new farming practices Farmers still have an affinity towards print media reports and agricultural pages and they have complaints to print media on the removal of agricultural information pages from news papers. They prefer the reports in print media as it facilitates them to collect and refer articles when they need it. Farmers are having an eye of doubt about the credibility of farm programs by private T.V. channels. Even if they prefer private Television channels for listening and adopting new farming methods and other farm information, they scrutinize programs to know whether they are sponsored programs by agrochemical or agro-fertilizer manufacturer.

Electronic structure and properties of Cu2O

The structural and electronic properties of Cu2O have been investigated using the periodic Hartree-Fock method and a posteriori density-functional corrections. The lattice parameter, bulk modulus, and elastic constants have been calculated. The electronic structure of and bonding in Cu2O are analyzed and compared with x-ray photoelectron spectroscopy spectra, showing a good agreement for the valence-band states. To check the quality of the calculated electron density, static structure factors and Compton profiles have been calculated, showing a good agreement with the available experimental data. The effective electron and hole masses have been evaluated for Cu2O at the center of the Brillouin zone. The calculated interaction energy between the two interpenetrated frameworks in the cuprite structure is estimated to be around -6.0 kcal/mol per Cu2O formula. The bonding between the two independent frameworks has been analyzed using a bimolecular model and the results indicate an important role of d10-d10 type interactions between copper atoms.

Implementation and Evaluation of an Improved Header Compression for 6LoWPAN

Extending IPv6 to IEEE 802.15.4-based Low power Wireless Personal Area Networks requires efficient header compression mechanisms to adapt to their limited bandwidth, memory and energy constraints. This paper presents an experimental evaluation of an improved header compression scheme which provides better compression of IPv6 multicast addresses and UDP port numbers compared to existing mechanisms. This scheme outperforms the existing compression mechanism in terms of data throughput of the network and energy consumption of nodes. It enhances throughput by up to 8% and reduces transmission energy of nodes by about 5%.

Analysis of the electronic structure, hyperfine structure, and volume isotope shifts in the low lying states of BA I and BA II

Relativistic multi-configuration Dirac Fock (MCDF) wavefunctions coupled to good angular momentum J have been calculated for low lying states of Ba I and Ba II. These wavefunctions are compared with semiempirical ones derived from experimental atomic energy levels. It is found that significantly better agreement is obtained when close configurations are included in the MCDF wavefunctions. Calculations of the electronic part of the field isotope shift lead to very good agreement with electronic factors derived from experimental data. Furthermore, the slopes of the lines in a King plot analysis of many of the optical lines are predicted accurately by these calculations. However, the MCDF wavefunctions seem not to be of sufficient accuracy to give agreement with the experimental magnetic dipole and electric quadrupole hyperfine structure constants.

Electronic structure and properties of the group 4, 5 and 6 highest chlorides including elements 104, 105, and 106

Results of the Dirac-Slater discrete variational calculations for the group 4, 5, and 6 highest chlorides including elements 104, 105, and 106 have shown that the groups are not identical with respect to trends in the electronic structure and bonding. The charge density distribution data show that notwithstanding the basic increase in covalency within the groups this increase diminishes in going from group 4 to group 6. As a result, E106Cl_6 will be less stable toward thermal decomposition than WCl_6, which is confirmed by an estimated low E106-Cl bond energy. \delta H_form equal to -90.3 ± 6 kcal/rnol is obtained for E106Cl_6 in the gas phase, which is indicative of a very low stability of this compound. The stability of the maximum oxidation state is shown to decrease in the direction E104(+4) > E105(+5) > E106(+6).

Microscale thermal engineering of electronic systems

The electronics industry is encountering thermal challenges and opportunities with lengthscales comparable to or much less than one micrometer. Examples include nanoscale phonon hotspots in transistors and the increasing temperature rise in onchip interconnects. Millimeter-scale hotspots on microprocessors, resulting from varying rates of power consumption, are being addressed using two-phase microchannel heat sinks. Nanoscale thermal data storage technology has received much attention recently. This paper provides an overview of these topics with a focus on related research at Stanford University.

Financial Information Mediation: A Case Study of Standards Integration for Electronic Bill Presentment and Payment Using the COIN Mediation Technology

Each player in the financial industry, each bank, stock exchange, government agency, or insurance company operates its own financial information system or systems. By its very nature, financial information, like the money that it represents, changes hands. Therefore the interoperation of financial information systems is the cornerstone of the financial services they support. E-services frameworks such as web services are an unprecedented opportunity for the flexible interoperation of financial systems. Naturally the critical economic role and the complexity of financial information led to the development of various standards. Yet standards alone are not the panacea: different groups of players use different standards or different interpretations of the same standard. We believe that the solution lies in the convergence of flexible E-services such as web-services and semantically rich meta-data as promised by the semantic Web; then a mediation architecture can be used for the documentation, identification, and resolution of semantic conflicts arising from the interoperation of heterogeneous financial services. In this paper we illustrate the nature of the problem in the Electronic Bill Presentment and Payment (EBPP) industry and the viability of the solution we propose. We describe and analyze the integration of services using four different formats: the IFX, OFX and SWIFT standards, and an example proprietary format. To accomplish this integration we use the COntext INterchange (COIN) framework. The COIN architecture leverages a model of sources and receivers’ contexts in reference to a rich domain model or ontology for the description and resolution of semantic heterogeneity.

CODAPACK3D. A new version of Compositional Data Package

The statistical analysis of compositional data should be treated using logratios of parts, which are difficult to use correctly in standard statistical packages. For this reason a freeware package, named CoDaPack was created. This software implements most of the basic statistical methods suitable for compositional data. In this paper we describe the new version of the package that now is called CoDaPack3D. It is developed in Visual Basic for applications (associated with Excel©), Visual Basic and Open GL, and it is oriented towards users with a minimum knowledge of computers with the aim at being simple and easy to use. This new version includes new graphical output in 2D and 3D. These outputs could be zoomed and, in 3D, rotated. Also a customization menu is included and outputs could be saved in jpeg format. Also this new version includes an interactive help and all dialog windows have been improved in order to facilitate its use. To use CoDaPack one has to access Excel© and introduce the data in a standard spreadsheet. These should be organized as a matrix where Excel© rows correspond to the observations and columns to the parts. The user executes macros that return numerical or graphical results. There are two kinds of numerical results: new variables and descriptive statistics, and both appear on the same sheet. Graphical output appears in independent windows. In the present version there are 8 menus, with a total of 38 submenus which, after some dialogue, directly call the corresponding macro. The dialogues ask the user to input variables and further parameters needed, as well as where to put these results. The web site http://ima.udg.es/CoDaPack contains this freeware package and only Microsoft Excel© under Microsoft Windows© is required to run the software. Kew words: Compositional data Analysis, Software

Privacy and Electronic Communications

slides for a class which explores professional, ethical and legal issues surround the use, storage and transmission of electronic communications and data. Follows on from previous class which looked in greater detail at the Data Protection Act

Data-Driven Text Generation using Neural Networks & Provenance is Complicated and Boring — Is there a solution?

Title: Data-Driven Text Generation using Neural Networks Speaker: Pavlos Vougiouklis, University of Southampton Abstract: Recent work on neural networks shows their great potential at tackling a wide variety of Natural Language Processing (NLP) tasks. This talk will focus on the Natural Language Generation (NLG) problem and, more specifically, on the extend to which neural network language models could be employed for context-sensitive and data-driven text generation. In addition, a neural network architecture for response generation in social media along with the training methods that enable it to capture contextual information and effectively participate in public conversations will be discussed. Speaker Bio: Pavlos Vougiouklis obtained his 5-year Diploma in Electrical and Computer Engineering from the Aristotle University of Thessaloniki in 2013. He was awarded an MSc degree in Software Engineering from the University of Southampton in 2014. In 2015, he joined the Web and Internet Science (WAIS) research group of the University of Southampton and he is currently working towards the acquisition of his PhD degree in the field of Neural Network Approaches for Natural Language Processing. Title: Provenance is Complicated and Boring — Is there a solution? Speaker: Darren Richardson, University of Southampton Abstract: Paper trails, auditing, and accountability — arguably not the sexiest terms in computer science. But then you discover that you've possibly been eating horse-meat, and the importance of provenance becomes almost palpable. Having accepted that we should be creating provenance-enabled systems, the challenge of then communicating that provenance to casual users is not trivial: users should not have to have a detailed working knowledge of your system, and they certainly shouldn't be expected to understand the data model. So how, then, do you give users an insight into the provenance, without having to build a bespoke system for each and every different provenance installation? Speaker Bio: Darren is a final year Computer Science PhD student. He completed his undergraduate degree in Electronic Engineering at Southampton in 2012.

Data collection in Internet environment in social representations studies

The present study aimed at comparing social representations structures concerning data collection procedures: through internet forms, diffused in the WWW, and through conventional paper and pencil questionnaire methods. overall 893 individuals participated in the research, 58% of whom were female. A total of 217 questionnaires about the social representation on football (soccer) and 218 about the representation on aging were answered by Brazilian university students in classrooms. Electronic versions of the same instrument were diffused through an internet forum linked to the same university. There were 238 answers for the football questionnaire and 230 for the aging one. The instrument asked participants to indicate five words or expressions related to one of the social objects. Sample characteristics and structural analyses were carried out separately for the two data collection procedures. data indicated that internet-based research allows for higher sample diversity, but it is essential to guarantee the adoption of measures that can select only desired participants. Results also pointed out the need to take into account the nature of the social object to be investigated through internet research on representations, seeking to avoid self-selection effects, which can bias results, as it seems to have happened with the football social object.

Impacto del Internet en el ingreso laboral: análisis del lugar y manejo del Internet

Este estudio analiza el efecto del uso de Internet sobre el ingreso laboral, eliminado el problema de endogeneidad que existe por motivos de selección. También se analizan los efectos seg un el lugar de acceso y las actividades realizadas en Internet. Para esto se utiliza Propensity Score Matching, tomando información para Colombia entre el año 2009 a 2011. Los resultados muestran que existe un efecto positivo, significativo y heterogéneo entre trabajadores asalariados y cuenta propia, siendo los últimos los más beneficiados. Con respecto al lugar de uso se encuentra un impacto positivo del uso del Internet en el trabajo y/o en el hogar, con una mayor magnitud en el primer sitio de acceso. Para el tema del manejo del Internet, no se encuentra efecto cuando se utiliza en actividades de entretenimiento, pero si cuando se usa en tareas educativas diariamente. Esto siguiere que usar este servicio aumenta la productividad y el capital humano, lo que se transforma en un incremento en los ingresos. Por tanto, se encuentra que el lugar de uso es una señal informativa para el mercado laboral, mientras que el impacto del manejo de esta herramienta depende de la frecuencia con que se realicen las actividades.

Estimation of electronic coupling in π -stacked donor-bridge-acceptor systems: correction of the two-state model

Comparison of donor-acceptor electronic couplings calculated within two-state and three-state models suggests that the two-state treatment can provide unreliable estimates of Vda because of neglecting the multistate effects. We show that in most cases accurate values of the electronic coupling in a π stack, where donor and acceptor are separated by a bridging unit, can be obtained as Ṽ da = (E2 - E1) μ12 Rda + (2 E3 - E1 - E2) 2 μ13 μ23 Rda2, where E1, E2, and E3 are adiabatic energies of the ground, charge-transfer, and bridge states, respectively, μij is the transition dipole moments between the states i and j, and Rda is the distance between the planes of donor and acceptor. In this expression based on the generalized Mulliken-Hush approach, the first term corresponds to the coupling derived within a two-state model, whereas the second term is the superexchange correction accounting for the bridge effect. The formula is extended to bridges consisting of several subunits. The influence of the donor-acceptor energy mismatch on the excess charge distribution, adiabatic dipole and transition moments, and electronic couplings is examined. A diagnostic is developed to determine whether the two-state approach can be applied. Based on numerical results, we showed that the superexchange correction considerably improves estimates of the donor-acceptor coupling derived within a two-state approach. In most cases when the two-state scheme fails, the formula gives reliable results which are in good agreement (within 5%) with the data of the three-state generalized Mulliken-Hush model