13197 research outputs found
Sort by
Natural Language Processing for Lower-Resource Code-Switching Languages: Case Studies on Transliteration and Resource Building for Tatar
Get PDF奈良先端科学技術大学院大学修士 (工学)master thesi
リング オシレータ ベース ノ シンセイ ランスウ セイセイキ ニ タイスル シュウハスウ チュウニュウ コウゲキ タイセイ ヒョウカ システム ノ テイアン
Get PDF奈良先端科学技術大学院大学修士(工学)master thesi
カネツ レイキャク カテイ ニ オケル ヒョウメン オンド ノ ケイジ ヘンカ ニ モトヅク ドジョウ ガンスイ ジョウタイ スイテイ
Get PDF奈良先端科学技術大学院大学修士(工学)master thesi
カイゴシ ノ フタン ケイゲン ニ ムケタ ストレス シヒョウ ニ モトヅク シンシン ジョウタイ ヘンカ ノ カシカ ト ブンセキ
Get PDF奈良先端科学技術大学院大学修士(工学)master thesi
ナラマチ ハッピー マップ : クラウド ソーシング ト シンソウ ガクシュウ ニ ヨル マチ ノ フンイキ スイテイ シュホウ ト ケイロ テイアン
Get PDF奈良先端科学技術大学院大学修士(工学)master thesi
Automated Quantization and Retraining for Neural Network Models Without Labeled Data
No full textDeploying neural network models to edge devices is becoming increasingly popular because such deployment decreases the response time and ensures better data privacy of services. However, running large models on edge devices poses challenges because of limited computing resources and storage space. Researchers have therefore proposed various model compression methods to reduce the model size. To balance the trade-off between model size and accuracy, conventional model compression methods require manual effort to find the optimal configuration that reduces the model size without significant degradation of accuracy. In this article, we propose a method to automatically find the optimal configurations for quantization. The proposed method suggests multiple compression configurations that produce models with different size and accuracy, from which users can select the configurations that suit their use cases. Additionally, we propose a retraining method that does not require any labeled datasets for retraining. We evaluated the proposed method using various neural network models for classification, regression and semantic similarity tasks, and demonstrated that the proposed method reduced the size of models by at least 30% while maintaining less than 1% loss of accuracy. We compared the proposed method with state-of-the-art automated compression methods, and showed that it can provide better compression configurations than existing methods.journal articl
Reflective action selection based on positive-unlabeled learning and causality detection model
No full textTask-oriented dialogue systems need to take appropriate actions not only for clear user requests but also for ambiguous and vague ones. In this study, “ambiguous” denotes that although users have potential requests, they failed to clearly define and verbalize their content and conditions which can be associated with system actions. For such ambiguous requests, taking reflective actions is one plausible choice for such systems. In our study, “reflective” denotes taking actions that satisfy user requests before the users themselves clarify their demands. We constructed such a reflective dialogue agent by collecting a corpus that includes pairs of ambiguous user requests and corresponding reflective system actions on sightseeing navigation with a smartphone. Since annotating every possible combination of user requests and system actions is impossible, this study built a corpus where one reflective action is annotated to one ambiguous user request. To train an action selection model on such incomplete training data in which only one action is associated with a request, we applied the positive/unlabeled (PU) learning method, which assumes that only part of the data is labeled with positive examples. In addition, we enhanced the action selection by extracting and distilling knowledge that corresponds to causality from the training data using a causality detection model. The experimental results show that both the PU learning method and the causality detection model improved the performances of the reflective action selection compared to the conventional positive/negative (PN) learning method.journal articl
Giving Back: Contributions Congruent to Library Dependency Changes in a Software Ecosystem
No full textThe widespread adoption of third-party libraries for contemporary software development has led to the creation of large inter-dependency networks, where sustainability issues of a single library can have widespread network effects. Maintainers of these libraries are often overworked, relying on the contributions of volunteers to sustain these libraries. To understand these contributions, in this work, we leverage socio-technical techniques to introduce and formalise dependency-contribution congruence (DC congruence) at both ecosystem and library level, i.e., to understand the degree and origins of contributions congruent to dependency changes, analyze whether they contribute to library dormancy (i.e., a lack of activity), and investigate similarities between these congruent contributions compared to typical contributions. We conduct a large-scale empirical study to measure the DC congruence for the npm ecosystem using 1.7 million issues, 970 thousand pull requests (PRs), and over 5.3 million commits belonging to 107,242 npm libraries. We find that the most congruent contributions originate from contributors who can only submit (not commit) to both a client and a library. At the project level, we find that DC congruence shares an inverse relationship with the likelihood that a library becomes dormant. Specifically, a library is less likely to become dormant if the contributions are congruent with upgrading dependencies. Finally, by comparing the source code of contributions, we find statistical differences in the file path and added lines in the source code of congruent contributions when compared to typical contributions. Our work has implications to encourage dependency contributions, especially to support library maintainers in sustaining their projects.journal articl