| 目黒 豊美 | 1361201: D, 中間発表 | 中村 哲，松本 裕治，戸田 智基，SAKRIANI SAKTI，GRAHAM NEUBIG |
|
title: Non-task-oriented dialogue system based on statistical model
abstract: Our aim is to build non-task-oriented dialoguesystembased on statistical model. In this presentation, we show (1) the analysis of listening-oriented dialogue(LoD) to reveal the characteristics of LoD, (2) a dialogue control method using POMDP for LoD, (3) dialogue act tagging for microblog utterances using semantic category patterns, and (4) the fusion of rule-based and stochastic utterance generation. language of the presentation: Japanese | ||
| LASGUIDO | 1361205: D, 中間発表 | 中村 哲，松本 裕治，戸田 智基，SAKRIANI SAKTI，GRAHAM NEUBIG |
|
Title:
An Example-Based Chat-Oriented Dialog Systems
Abstract: We present our work on developing a chat-oriented dialog system by utilizing real human-to-human conversation from various data source like movie script and Twitter conversation. The aim of the proposed method is to build a conversational agent that can interact with users in as natural a fashion as possible, while reducing the time requirement for database design and collection. A number of the challenging design issues we faced are described, including (1) constructing an appropriate dialog corpora from raw movie scripts and Twitter data, and (2) developing an multi domain chat-oriented dialog management system which can retrieve a proper system response based on the current user query. To build a dialog corpus, we propose a unit of conversation called a tri-turn (a trigram conversation turn), as well as extraction and semantic similarity analysis techniques to help ensure that the content extracted from raw movie/drama script files forms appropriate dialog-pair (query-response) examples. The constructed dialog corpora are then utilized in a data-driven dialog management system. Here, various approaches are investigated including example-based (EBDM) and response generation using phrase-based statistical machine translation (SMT). In particular, we use two EBDM: syntactic-semantic similarity retrieval and TF-IDF based cosine similarity retrieval. Experiments are conducted to compare and contrast EBDM and SMT approaches in building a chat-oriented dialog system, and we investigate a combined method that addresses the advantages and disadvantages of both approaches. Furthemore, we improve the response retrieval accuracy and robustness by utilizing recursive neural network paraphrase identification. We model our dialog-pair database and user input query with distributed word representations, and employ recursive autoencoders and dynamic pooling to determine whether two sentences with arbitrary length have the same meaning. The distributed representations have the potential to improve handling of out of vocabulary (OOV) cases, and the recursive structure can reduce confusion in example matching. System performance was evaluated based on objective metrics and human subjective metric. Experimental results show that the proposed filtering approach effectively improve the performance. The results also show that by combing both EBDM and SMT approaches, we could overcome the shortcomings of each. Furthermore, by utilizing recursive neural network paraphrase identification we manage to increase the system accuracy and robustness in handling OOV cases. Language of the presentation: English | ||
| 麻生 栄樹 | 1451005: M, 1回目発表 | 自然言語処理学 |
|
title: Automatic Generation of Structured Abstract for Science Documents
abstract: In the scene of research, we read a lot of papers for reference. Reading paper takes us long time, so we're eager for the way of shorten them. In my research, I propose an automatic summarizing system using structured abstract as training data. On this time, I will introduce the method of it and talk about progress of my research. language of the presentation: Japanese | ||
| 小松 巡 | 1451201: M, 1回目発表 | ディペンダブルシステム学 |
| AURELIO CORTESE | 1361203: D, 中間発表 | 池田 和司，小笠原 司，川人 光男 |
|
Title: Using multivoxel pattern neurofeedback to selectively manipulate subjective awareness without changing perceptual performance Abstract: One important aspect of conscious awareness is subjective confidence. For instance, in the classic neurological phenomenon of blindsight, where subjective awareness is impaired, it is confidence rather than perceptual capacity that is abolished; these patients can discriminate visual stimuli above chance, although they claim to be just guessing with little confidence. Recent work using fMRI in humans as well as neuronal recording in animals have identified correlates of perceptual confidence in frontal and parietal brain regions. However, one issue is whether such correlates are causally critical for confidence to arise. A yet more concerning issue is that confidence is typically confounded with perceptual performance, so that these correlates may just reflect internal visual signal strength rather than subjective confidence per se. Here we used a recently developed method of fMRI neurofeedback, called decoded neurofeedback (DecNef), to address these issues. We first used multivoxel pattern analysis technique (sparse logistic regression) to learn the spatial pattern of brain activity associated with subjective confidence. Participants were presented with moving dot stimuli near perceptual threshold, and they made responses to discriminate between motion directions before they rated their confidence in each trial. Based on the fMRI activity recorded while subjects were performing these tasks, we computationally constructed a “decoder” from prefrontal and parietal regions that reliably classified between high and low confidence states. In subsequent experimental sessions, subjects were given online feedback of this decoded information, such that with monetary reward cues they were trained to implicitly induce different confidence states by directly changing their own brain activity. As predicted, we found that these induced brain states indeed led to differences in subjective ratings of confidence, when subjects were presented with the same dot motion stimuli. This shows that these correlates of confidence in prefrontal and parietal areas are causally relevant. Importantly, self-induction of different confidence states based on DecNef was selective in that it left task performance unchanged. This ruled out the important confound that these correlates may just reflect motion discrimination capacity. These results demonstrate a novel method to robustly and non-invasively dissociate confidence and awareness from the basic mechanisms for perceptual decisions. Language of the presentation: English | ||
| CHRISTIAN DEUS TELMO CAYAO | 1351204: M, 2回目発表 | 小笠原 司，向川 康博，高松 淳，池田 篤俊 |
|
Title: Stereo Vision System with Water Surface Refraction-Correction for Robot Manipulation Abstract: Vision is an effective sensing form for humans and animals in understanding their environment, recognizing objects around them, and interacting with the environment. Stereo vision serves as a sensing unit for robots which can perceive depth using a pair of eyes like humans and animals. We propose robot vision for manipulating underwater objects. In this research, we use a fiducial marker to estimate the configuration of the water surface. We obtain epipolar curves, not epipolar lines, by simulating the refraction. We improve standard block-matching-based stereo using the curves and show the result of the stereo reconstruction of the underwater objects. Language of the presentation: English | ||
| LUIZ GUSTAVO MOREIRA SAMPAIO | 1351209: M, 2回目発表 | 加藤 博一，萩田 紀博，CHRISTIAN SANDOR，山本 豪志朗 |
|
title: Mobile System for Interaction with Wall-Sized Displays
abstract: We exploit the difference of perception between the human visual system and the camera sensors to embed binary patterns on off-the-shelf displays. These patterns are detectable by cameras running with specific frame rate while remaining imperceptible to observers. We explain how we use this technology with arbirtrary dynamic contents, such as movies and animations, targeting interaction with public huge displays. language of the presentation: English | ||
| PATHIRANNAHALAGE SHALIKA PRABHANI PATHIRATHNA | 1451206: M, 1回目発表 | インタラクティブメディア設計学 |
|
title: Effect of Alignment Errors of Video Guides on Manual Task Performance abstract: Video guides for manual tasks are easy to make and production cost is very low. Also they contain very useful information. But there is a disadvantage of manual task video guides. When the video is taken from one angle then the user can look at the scene from a different angle than the video. In this case there is a difference between what user see in the video guide and what he actually see. This might cause to task performance. Therefore we will conduct an user study to identify how video guides alignment errors will effect to performance of manual task. language of the presentation: English | ||