13197 research outputs found
Sort by
Modeling Unsupervised Empirical Adaptation by DPGMM and DPGMM-RNN Hybrid Model to Extract Perceptual Features for Low-Resource ASR
No full textSpeech feature extraction is critical for ASR systems. Such successful features as MFCC and PLP use filterbank techniques to model log-scaled speech perception but fail to model the adaptation of human speech perception by hearing experiences. Infant perception that is adapted by hearing speech without text may cause permanent brain state modifications (engrams) that serve as a physical fundamental basis for lifetime speech perception formation. This realization motivates us to propose to model such an unsupervised adaptation process, where adaptation denotes perception that is affected or changed by the history of experiences, with the Dirichlet Process Gaussian Mixture Model (DPGMM) and the DPGMM-RNN hybrid model to extract perceptual features to improve ASR. Our proposed features extend MFCC features with posteriorgrams extracted from the DPGMM algorithm or the DPGMM-RNN hybrid model. Our analysis shows that the DPGMM and DPGMM-RNN model perplexities agree with infant auditory perplexity to support that the proposed features are perceptual. Our ASR results verify the effectiveness of the proposed unsupervised features in such tasks as LVCSR on WSJ and ASR on noisy low-resource telephone conversations, compared with the supervised bottleneck features from Kaldi in ASR performance.journal articl
Automatic Machine Translation Evaluation using a Source and Reference Sentence with a Cross-lingual Language Model
No full textAs the performance of machine translation has improved, the need for a human-like automatic evaluation metric has been increasing. The use of multiple reference translations against a system translation (a hypothesis) has been adopted as a strategy to improve the performance of such evaluation metrics. However, preparing multiple references is highly expensive and impractical. In this study, we propose an automatic evaluation method for machine translation that uses source sentences as additional pseudo-references. The proposed method evaluates a translation hypothesis via regression to assign a real-valued score. The model takes the paired source, reference, and hypothesis sentences together as input. A pre-trained large-scale cross-lingual language model encodes the input to sentence vectors, with which the model predicts a human evaluation score. The results of experiments show that our proposed method exhibited stably higher correlation with human judgements than baseline methods that solely depend on hypothesis and reference sentences, especially when the hypotheses were very high- or low-quality translations.journal articl
オンラインアンケートにおける不適切回答自動検出に向けた回答操作ログの統計分析
Get PDFオンラインアンケートにおいて，なるべく楽に早くアンケートタスクを完了しようとする「Satisficing（努力の最小限化）」という態度が調査結果の信頼性を低下させる問題がある．この問題に対し，回答時間が明らかに短い回答を除外するなどの措置が考案されてきたが，回答時間だけでは完全に除外することができない．そこで我々は，回答時間だけでなく，回答中の画面操作を観測することで，より高精度にSatisficingが検出可能になるのではないかという仮説を立てた．世界中で利用されているオープンソースアンケートシステムLimeSurveyの拡張機能として，回答中の画面操作ログ（回答操作ログ）を記録できるプラグインを開発し，本システムを用いてクラウドソーシング上でアンケート実験を行った．実験は，回答者は各自のスマートフォンを用いて，リッカート形式と自由記述形式の質問から成るアンケートに回答してもらうものであり，アンケートの途中にはSatisficingを検出するために考案されている既存の質問を挿入した．1000人に対して回答を依頼し，収集した回答内容および回答操作ログから，Satisficingを表現する可能性が考えられる特徴量を生成し，特徴量ごとにSatisficingである群とSatisficingでない群の母平均の差の検定を行った．結果として，既存の特徴量であるテキストの文字数，連続同一回答数，中間回答数に加え，開発したプラグインでのみ取得可能なスクロール速度，選択肢の変更回数，リッカート形式の回答時間などに統計的有意差があることを明らかにした．また，スクロール速度やテキストの変更回数に関しては，絶対値では差がない一方で，回答者ごとのベースラインとの差では有意な差が確認できた．journal articl
ParmoSense: Scenario-based Participatory Mobile Urban Sensing Platform with User Motivation Engine
No full textThe rapid proliferation of mobile devices with various sensors has enabled participatory mobile sensing (PMS). Several PMS platforms suffer from open issues including the limited use of their functions to a specific scenario/case and the necessity of technical knowledge for organizers. This paper proposes a novel PMS platform named ParmoSense for easy and flexible data collection. To reduce the burden on both organizers and participants, we employ two novel features. First, essential PMS functions implemented as modules can be easily chosen and combined for sensing in different scenarios. Second, the scenario-based description feature allows organizers to easily and quickly prepare a new instance of PMS and enable people to easily participate in the sensing. It also provides multiple functions to motivate participants for sustainable operation. Through a performance comparison with existing PMS platforms, we confirmed that ParmoSense shows the best cost performance in terms of the workload for preparation and the variety of functions. In addition, to evaluate the availability and usability of ParmoSense, we conducted 19 case studies over four years with ordinary citizens. As the result of a questionnaire survey carried out during the case studies, we confirmed that ParmoSense can be easily operated by ordinary citizens without technical skills.journal articl
Iterative Span Selection: Self-Emergence of Resolving Orders in Semantic Role Labeling
No full textSemantic Role Labeling (SRL) is the task of labeling semantic arguments for marked semantic predicates. Semantic arguments and their predicates are related in various distinct manners, of which certain semantic arguments are a necessity while others serve as an auxiliary to their predicates. To consider such roles and relations of the arguments in the labeling order, we introduce iterative argument identification (IAI), which combines global decoding and iterative identification for the semantic arguments. In experiments, we first realize that the model with random argument labeling orders outperforms other heuristic orders such as the conventional left-to-right labeling order. Combined with simple reinforcement learning, the proposed model spontaneously learns the optimized labeling orders that are different from existing heuristic orders. The proposed model with the IAI algorithm achieves competitive or outperforming results from the existing models in the standard benchmark datasets of span-based SRL: CoNLL-2005 and CoNLL-2012.conference pape
Uncertainty-Aware Manipulation Planning Using Gravity and Environment Geometry
No full textFactory automation robot systems often depend on specially-made jigs that precisely position each part, which increases the system's cost and limits flexibility. We propose a method to determine the 3D pose of an object with high precision and confidence, using only parallel robotic grippers and no parts-specific jigs. Our method automatically generates a sequence of actions that ensures that the real-world position of the physical object matches the system's assumed pose to sub-mm precision. Furthermore, we propose the use of “extrinsic” actions, which use gravity, the environment and the gripper geometry to significantly reduce or even eliminate the uncertainty about an object's pose. We show in simulated and real-robot experiments that our method outperforms our previous work, at success rates over 95%.journal articl
Independent Set Reconfiguration on Directed Graphs
No full textDirected Token Sliding asks, given a directed graph and two sets of pairwise nonadjacent vertices, whether one can reach from one set to the other by repeatedly applying a local operation that exchanges a vertex in the current set with one of its out-neighbors, while keeping the nonadjacency. It can be seen as a reconfiguration process where a token is placed on each vertex in the current
set, and the local operation slides a token along an arc respecting its direction. Previously, such a problem was extensively studied on undirected graphs, where the edges have no directions and thus the local operation is symmetric. Directed Token Sliding is a generalization of its undirected variant since an undirected edge can be simulated by two arcs of opposite directions. In this paper, we initiate the algorithmic study of Directed Token Sliding. We first observe that the problem is PSPACE-complete even if we forbid parallel arcs in opposite directions and that the problem on directed acyclic graphs is NP-complete and W[1]-hard parameterized by the size of the sets in consideration. We then show our main result: a linear-time algorithm for the problem on directed graphs whose underlying undirected graphs are trees, which are called polytrees. Such a result is also known for the undirected variant of the problem on trees [Demaine et al. TCS 2015], but the techniques used here are quite different because of the asymmetric nature of the directed problem. We present a characterization of yes-instances based on the existence of a certain set of directed paths, and then derive simple equivalent conditions from it by some observations, which yield an efficient algorithm. For the polytree case, we also present a quadratic-time algorithm that
outputs, if the input is a yes-instance, one of the shortest reconfiguration sequences.conference pape
シロイヌナズナ ニ オケル ネ カンセン シジョウキン ノ カンセン ヨウシキ ト ビョウゲンセイ ハツゲン ノ カンケイセイ ニ ツイテ
Get PDF奈良先端科学技術大学院大学修士(バイオサイエンス)master thesi
マウス イ ジョウヒ サイボウ オルガノイド オ モチイタ ピロリキン ビョウゲン インシ CagA ノ キノウ カイセキ
Get PDF奈良先端科学技術大学院大学修士(バイオサイエンス)master thesi
A Large Scale Study on the Synchronicity of Libraries that cross Different Ecosystems
Get PDF奈良先端科学技術大学院大学修士(工学)master thesi