The Experts below are selected from a list of 112266 Experts worldwide ranked by ideXlab platform

Daniel Joyce - One of the best experts on this subject based on the ideXlab platform.

Steven C. H. Hoi - One of the best experts on this subject based on the ideXlab platform.

  • Video-Grounded Dialogues with Pretrained Generation Language Models.
    arXiv: Computation and Language, 2020
    Co-Authors: Steven C. H. Hoi
    Abstract:

    Pre-trained Language models have shown remarkable success in improving various downstream NLP tasks due to their ability to capture dependencies in textual data and generate natural responses. In this paper, we leverage the power of pre-trained Language models for improving video-grounded dialogue, which is very challenging and involves complex features of different dynamics: (1) Video features which can extend across both spatial and temporal dimensions; and (2) Dialogue features which involve semantic dependencies over multiple dialogue turns. We propose a framework by extending GPT-2 models to tackle these challenges by formulating video-grounded dialogue tasks as a sequence-to-sequence task, combining both visual and textual representation into a structured sequence, and fine-tuning a large pre-trained GPT-2 network. Our framework allows fine-tuning Language models to capture dependencies across multiple modalities over different levels of information: spatio-temporal level in video and token-sentence level in dialogue context. We achieve promising improvement on the Audio-Visual Scene-Aware Dialogues (AVSD) benchmark from DSTC7, which supports a potential direction in this line of research.

  • ACL - Video-Grounded Dialogues with Pretrained Generation Language Models
    Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020
    Co-Authors: Steven C. H. Hoi
    Abstract:

    Pre-trained Language models have shown remarkable success in improving various downstream NLP tasks due to their ability to capture dependencies in textual data and generate natural responses. In this paper, we leverage the power of pre-trained Language models for improving video-grounded dialogue, which is very challenging and involves complex features of different dynamics: (1) Video features which can extend across both spatial and temporal dimensions; and (2) Dialogue features which involve semantic dependencies over multiple dialogue turns. We propose a framework by extending GPT-2 models to tackle these challenges by formulating video-grounded dialogue tasks as a sequence-to-sequence task, combining both visual and textual representation into a structured sequence, and fine-tuning a large pre-trained GPT-2 network. Our framework allows fine-tuning Language models to capture dependencies across multiple modalities over different levels of information: spatio-temporal level in video and token-sentence level in dialogue context. We achieve promising improvement on the Audio-Visual Scene-Aware Dialogues (AVSD) benchmark from DSTC7, which supports a potential direction in this line of research.

Ann C. Wintergerst - One of the best experts on this subject based on the ideXlab platform.

  • Second-Generation Language Maintenance and Identity: A Case Study
    Bilingual Research Journal, 2009
    Co-Authors: Andrea Decapua, Ann C. Wintergerst
    Abstract:

    While the benefits of bilingualism have been widely acknowledged, parents face many hurdles raising children bilingually. Factors such as consistency in Language use, family, school, and social support networks, issues of ethnic and social identity, and the prestige value of Language have contributed to successful bilingualism. This paper presents a case study exploring the maintenance of German in an English-dominant environment, the strategies the mother employed in fostering German, and how her strategies influenced the children's perceptions of German. The findings offer insights into nurturing bilingualism, particularly when community and school do not support the heritage Language.

George Papamargaritis - One of the best experts on this subject based on the ideXlab platform.

  • aspects and constraints for implementing configurable product line architectures
    IEEE IFIP International Conference on Software Architecture, 2004
    Co-Authors: David Lesaint, George Papamargaritis
    Abstract:

    Component-based product-line architectures (PLAs) must support two operations: application configuration - the construction of valid application specifications - and application Generation - the compilation of specifications into executable applications. Whereas configuration is a combinatorial task involving advanced knowledge-based reasoning, Generation is a deterministic compilation process. This suggests an application synthesis model where configuration and Generation are carried out separately by interoperable tools. To this end, we introduce a PLA development toolkit which includes a constraint-based configuration Language and an aspect-based Generation Language supporting the same architecture model. The toolkit imposes dual PLA implementations consisting of a configuration program and a Generation program. The compilation of the configuration program yields an interactive configurator used to produce valid configurations at run-time. Valid configurations are then compiled by the generator with the Generation program to produce Java applications. Overall, this model allows the use of powerful configuration and Generation technologies - namely, constraint programming and aspect-oriented programming - while enforcing view consistency and tool interoperability.

  • WICSA - Aspects and constraints for implementing configurable product-line architectures
    Proceedings. Fourth Working IEEE IFIP Conference on Software Architecture (WICSA 2004), 2004
    Co-Authors: David Lesaint, George Papamargaritis
    Abstract:

    Component-based product-line architectures (PLAs) must support two operations: application configuration - the construction of valid application specifications - and application Generation - the compilation of specifications into executable applications. Whereas configuration is a combinatorial task involving advanced knowledge-based reasoning, Generation is a deterministic compilation process. This suggests an application synthesis model where configuration and Generation are carried out separately by interoperable tools. To this end, we introduce a PLA development toolkit which includes a constraint-based configuration Language and an aspect-based Generation Language supporting the same architecture model. The toolkit imposes dual PLA implementations consisting of a configuration program and a Generation program. The compilation of the configuration program yields an interactive configurator used to produce valid configurations at run-time. Valid configurations are then compiled by the generator with the Generation program to produce Java applications. Overall, this model allows the use of powerful configuration and Generation technologies - namely, constraint programming and aspect-oriented programming - while enforcing view consistency and tool interoperability.

  • Towards Software Product-Lines — A Refinement-Oriented Generation Language
    BT Technology Journal, 2003
    Co-Authors: David Lesaint, George Papamargaritis
    Abstract:

    With the advent of eBusiness and mobile computing, the need for software systems that can be automatically configured, assembled, and adapted on the fly has never been so critical. In this context, various proposals have been put forward to realise the vision of software product-lines, among them, GenVoca — a powerful model for component-based product-lines advocating large-scale step-wise refinement as a composition principle. In this paper, we introduce a refinement-oriented Generation Language — ReGaL — to program GenVoca product-lines. While components are programmed in Java, refinements are programmed in ReGaL by the means of generic refinement aspects. Orders for applications, themselves expressed in ReGaL, are compiled by instantiating and weaving refinement aspects with components to generate the requested Java applications. We illustrate ReGaL on a product-line of graph algorithms and present experiments showing the benefits of application configurability.

Paul S Bender - One of the best experts on this subject based on the ideXlab platform.

  • Large scale modeling for corporate planning
    Omega-international Journal of Management Science, 2003
    Co-Authors: Paul S Bender
    Abstract:

    The purpose of this paper is to outline the characteristics, use and benefits of a resource allocation system using a model Generation Language, and mixed integer programming. The system has been successfully implemented and used for over three years in a multi billion dollar international paper company, to support a wide variety of corporate planning needs.