担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The recognition of complex and subtle human behaviors from wearable sensors will enable next-generation human-oriented computing in scenarios of high societal value (e.g., dementia care). This will require large-scale human activity corpus and much improved methods to recognize activities and the context in which they occur. This workshop deals with the challenges of designing reproducible experimental setups, running large-scale dataset collection campaigns, designing activity and context recognition methods that are robust and adaptive, and evaluating systems in the real world. We wish to reflect on future methods, such as lifelong learning approaches that allow open-ended activity recognition.

DOI： 10.1145/3544793.3560377

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85158983039&origin=inward

担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：筆頭著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：英語   掲載種別：研究論文（学術雑誌）

In this study, we propose a simple and effective preprocessing method for subword segmentation based on a data compression algorithm. Compression-based subword segmentation has recently attracted significant attention as a preprocessing method for training data in neural machine translation. Among them, BPE/BPE-dropout is one of the fastest and most effective methods compared to conventional approaches; however, compression-based approaches have a drawback in that generating multiple segmentations is difficult due to the determinism. To overcome this difficulty, we focus on a stochastic string algorithm, called locally consistent parsing (LCP), that has been applied to achieve optimum compression. Employing the stochastic parsing mechanism of LCP, we propose LCP-dropout for multiple subword segmentation that improves BPE/BPE-dropout, and we show that it outperforms various baselines in learning from especially small training data.

DOI： 10.3390/electronics11071014

DOI： 10.3390/electronics11071014

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85126910361&origin=inward

担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：筆頭著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：筆頭著者,　責任著者   記述言語：日本語   掲載種別：研究論文（学術雑誌）

DOI： 10.11517/jsaisigtwo.2021.aimed-011_05

CiNii Research

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this paper we summarize the contributions of participants to the fourth Sussex-Huawei Locomotion-Transportation (SHL) Recognition Challenge organized at the HASCA Workshop of UbiComp/ISWC 2021. The goal of this machine learning/data science challenge is to recognize eight locomotion and transportation activities (Still, Walk, Run, Bike, Bus, Car, Train, Subway) from the radio sensor data (GPS location, GPS reception, WiFi reception and Cell reception) of a smartphone in a user-independent manner. The training and testing data are collected by different users with a smartphone placed at the Hips position. We introduce the dataset used in the challenge and the protocol of the competition. We present a meta-Analysis of the contributions from 15 submissions, their approaches, the software tools used, computational cost and the achieved results. The challenge evaluates the recognition performance by comparing predicted to ground-Truth labels at every second, but puts no constraints on the maximum decision window length. Overall, two submissions achieved F1 scores between 70% and 80%, one between 60% and 70%, five between 50% and 60%, and seven below 50%. Due to the technical challenges of data synchronization, sensor unavailability and sensor diversity, the overall performance based on GPS and radio sensors is lower than the performance achieved by motion sensors in previous challenges (SHL 2018-2020). Finally, we present a baseline implementation to help understand the contribution of each sensor modality to the recognition task.

DOI： 10.1145/3460418.3479373

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85115916446&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

The Sussex-Huawei Locomotion-Transportation (SHL) Recognition Challenges aim to advance and capture the state-of-the-art in locomotion and transportation mode recognition from smartphone motion (inertial) sensors. The goal of this series of machine learning and data science challenges was to recognize eight locomotion and transportation activities (Still, Walk, Run, Bus, Car, Train, Subway). The three challenges focused on time-independent (SHL 2018), position-independent (SHL 2019) and user-independent (SHL 2020) evaluations, respectively. Overall, we received 48 submissions (out of 93 teams who registered interest) involving 201 scientists over the three years. The survey captures the state-of-the-art through a meta-analysis of the contributions to the three challenges, including approaches, recognition performance, computational requirements, software tools and frameworks used. It was shown that state-of-the-art methods can distinguish with relative ease most modes of transportation, although the differentiating between subtly distinct activities, such as rail transport (Train and Subway) and road transport (Bus and Car) still remains challenging. We summarize insightful methods from participants that could be employed to address practical challenges of transportation mode recognition, for instance, to tackle over-fitting, to employ robust representations, to exploit data augmentation, and to exploit smart post-processing techniques to improve performance. Finally, we present baseline results to compare the three challenges with a unified recognition pipeline and decision window length.

DOI： 10.3389/fcomp.2021.713719

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85115915802&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Sensor-based human activity recognition technology has been used for estimating human action based on the sensor data. In this paper, we propose a new paradigm to render the human activity on a screen instead of classifying the activity among the activity labels. We could built this mockup of a simulator, combining our previous translation tool between signals [2] with the motion rendering systems [3]. We faced two problems which decrease the simulation ability a lot. We proposed two algorithms to increase the performance of this simulator in this preliminary work.

DOI： 10.1145/3460418.3479275

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85115964745&origin=inward

担当区分：責任著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

From the side of upper-level applications which require planning the actions in robot or those which need to search the whole log of activities in smart home, the action predicate expressions in the form of knowledge graphs may play an important role. The sequence of activities alone, which can be supplied by the conventional activity recognition systems, may not be sufficient for those applications. The subject of the particular activity is crucial information in most of the cases, and the object of the particular activity is often necessary to identify the characteristics. From this perspective, we have investigated the activities recognized by activity recognition systems, trying to identify their hidden elements which play the role of the subject and the object of the activities, i.e. activity knowledge graph. If we focus on these hidden elements, they are categorized in two: (1) person (subject)-person (object) interaction, and (2) person (subject)-object (object) interactions. Depending on the class of activities, these two are sometimes faced great difficulties: The hidden elements for walk, pick-up, open, and drink are quite easy but those for look-At, see, watch, and throw are difficult. The source of difficulties arises from the fact that the object (object) is not contacted from the person (subject). In this paper we have developed a method which identifies non-contacted object by the direction of the eye gaze of the person (subject) in the category of watch (activity). Using "Watching TV"data by Stair lab, the proposed system achieved 85% in accuracy.

DOI： 10.1145/3460418.3479351

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85115984916&origin=inward

担当区分：責任著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The recognition of complex and subtle human behaviors from wearable sensors will enable next-generation human-oriented computing in scenarios of high societal value (e.g., dementia care). This will require large-scale human activity corpus and much improved methods to recognize activities and the context in which they occur. This workshop deals with the challenges of designing reproducible experimental setups, running large-scale dataset collection campaigns, designing activity and context recognition methods that are robust and adaptive, and evaluating systems in the real world. We wish to reflect on future methods, such as lifelong learning approaches that allow open-ended activity recognition. This year HASCA will welcome papers from participants to the Fourth Sussex-Huawei Locomotion and Transportation Recognition Challenge and the Third Nursing Activity Recognition Challenge in special sessions.

DOI： 10.1145/3460418.3479266

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85115942696&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

DOI： 10.1145/3460418.3479391

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85115908587&origin=inward

担当区分：最終著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：責任著者   記述言語：日本語   掲載種別：研究論文（学術雑誌）

arXiv

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

While Human Action Recognition (HAR) using motion capture can perform well with high accuracy, it requires a high computational cost for recording and post-processing. To avoid this, we build a HAR system using 3D pose estimation from the single-camera video instead of motion capture. One drawback in this approach is that the performance is considerably dependent on the camera position. This paper investigates how we can use the pose estimate constantly without the effect of camera position even when the camera position in the test data is changed. We augment the data by rotating around the 3D pose estimate to improve the accuracy when using different camera positions in the test data and in the training data. The strategy of augmenting training data shows improvements up to 55.7% in accuracy, compared with the case of 2D pose with no augmentation.

DOI： 10.1007/978-981-15-8944-7_2

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85107353674&origin=inward

記述言語：英語   掲載種別：記事・総説・解説・論説等（学術雑誌）

DOI： 10.11517/jjsai.36.4_527

CiNii Article

その他リンク： https://ci.nii.ac.jp/naid/130008060910

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Action knowledge graphs can play a central role in smart cities, smart homes, robot planning, and so on. This is since both of the subject and the object of the actions can add more meaningful information for the higher-level application than the action alone as a predicate. We built a system that generates the action knowledge graphs from video using deep learning. Especially, we propose an algorithm which perceives the interaction between hand and object by measuring the proximity between them with considering the direction of fingers. We showed that this approach achieves the performance of 83% in accuracy using the Stair Lab data.

DOI： 10.1145/3410530.3414363

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85091839586&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this paper we summarize the contributions of participants to the third Sussex-Huawei Locomotion-Transportation (SHL) Recognition Challenge organized at the HASCA Workshop of UbiComp/ISWC 2020. The goal of this machine learning/data science challenge is to recognize eight locomotion and transportation activities (Still, Walk, Run, Bike, Bus, Car, Train, Subway) from the inertial sensor data of a smartphone in a user-independent manner with an unknown target phone position. The training data of a "train"user is available from smartphones placed at four body positions (Hand, Torso, Bag and Hips). The testing data originates from "test"users with a smartphone placed at one, but unknown, body position. We introduce the dataset used in the challenge and the protocol of the competition. We present a meta-analysis of the contributions from 15 submissions, their approaches, the software tools used, computational cost and the achieved results. Overall, one submission achieved F1 scores above 80%, three with F1 scores between 70% and 80%, seven between 50% and 70%, and four below 50%, with a latency of maximum of 5 seconds.

DOI： 10.1145/3410530.3414341

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85091885062&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

2nd Nurse Care Activity Recognition Challenge using Lab and Field Data is organized as a part of HASCA workshop and is continuation of Nurse Care Activity Recognition Challenge [7]. We give the description of the dataset and summarize the approaches used by the teams in this Challenge. In this challenge, data collected in both lab and real-world setting is provided to the challenge participants with an aim to bridge the gap between lab and practical field to reduce the workload of the nurses. The challenge was started on May 1, 2020 and continued until July 9, 2020. Accuracy is used as performance metric to evaluate the submissions. The winning team used k-NN classifier and achieved about 22.35% accuracy.

DOI： 10.1145/3410530.3414611

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85091886199&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The recognition of complex and subtle human behaviors from wearable sensors will enable next-generation human-oriented computing in scenarios of high societal value (e.g., dementia care). This will require large-scale human activity corpus and much improved methods to recognize activities and the context in which they occur. This workshop deals with the challenges of designing reproducible experimental setups, running large-scale dataset collection campaigns, designing activity and context recognition methods that are robust and adaptive, and evaluating systems in the real world. We wish to reflect on future methods, such as lifelong learning approaches that allow open-ended activity recognition. This year HASCA will welcome papers from participants to the Third Sussex-Huawei Locomotion and Transportation Recognition Challenge and Second Nursing Activity Recognition Challenge in special sessions.

DOI： 10.1145/3410530.3414612

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85091833400&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（学術雑誌）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

Virtual   2020年06月30日  -  2020年07月02日

担当区分：責任著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

担当区分：責任著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：英語   掲載種別：記事・総説・解説・論説等（学術雑誌）

DOI： 10.11517/jjsai.35.3_465

CiNii Article

その他リンク： https://ci.nii.ac.jp/naid/130007917827

担当区分：筆頭著者   記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：英語   掲載種別：研究論文（学術雑誌）

Human Activity Recognition from accelerometer sensors is key to enable applications such as fitness tracking or health status monitoring at home. However, evaluating the performance of activity recognition systems in real-life deployments is challenging to the multiple differences in sensor number, placement and orientation that may arise in real settings. Considering such differences requires a large amount of labeled data. To overcome the challenges and costs associated to the collection of a wide range of heterogeneous data, we propose a simulator, called MEASURed, which uses motion capture to simulate accelerometer data on different settings. Then, using the simulated data to estimate the performance of activity recognition models under different scenarios. In this chapter, we describe MEASURed and evaluate its performance in estimating the accuracy of activity recognition models. Our results show that MEASURed can estimate the average accuracy of an activity recognition model using real accelerometer magnitude data. By using motion capture to simulate accelerometer data, the sensor research community can profit from visual datasets that have been collected by other communities to evaluate performance of activity recognition in a wide range of activities. MEASURed can be used to evaluate activity recognition classifiers in settings with different number, placement, and sampling rate of accelerometer sensors. The evaluation on a broad spectrum of scenarios gives a more general view of models and their limitations.

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this paper we summarize the contributions of participants to the Sussex-Huawei Transportation-Locomotion (SHL) Recognition Challenge organized at the HASCA Workshop of UbiComp 2019. The goal of this machine learning/data science challenge is to recognize eight locomotion and transportation activities (Still, Walk, Run, Bike, Bus, Car, Train, Subway) from the inertial sensor data of a smartphone in a placement independent manner. The training data is collected with smartphones placed at three body positions (Torso, Bag and Hips), while the testing data is collected with a smartphone placed at another body position (Hand). We introduce the dataset used in the challenge and the protocol for the competition. We present a meta-analysis of the contributions from 14 submissions, their approaches, the software tools used, computational cost and the achieved results. Overall, three submissions achieved F1 scores between 70% and 80%, five with F1 scores between 60% and 70%, five between between 50% and 60%, and one below 50%, with a latency of a maximum of 5 seconds.

DOI： 10.1145/3341162.3344872

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85072892002&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this paper, activity recognition is performed using an optical motion capture system that can measure three-dimensional position information of reflective markers attached to the body. The individual markers detected by motion capture are automatically associated with which part of the body they are attached to. However, due to the overlapping of obstacles and other body parts and misplacement of the markers, these may be hidden from the camera and enter a blind spot, which may frequently cause a marker to be associated to different body parts erroneously. Usually, these errors need to be corrected manually after measurement, but this work is very time consuming, cumbersome and requires some skill. In this research, it is thought that there is no problem in recognizing the activity even if the process of spending the effort of correcting the correspondence between the marker after measurement and the body is omitted in the activity recognition using the motion capture. Because feature quantities are extracted from activity data when performing action recognition, even if an error occurs in part of the marker data, the effect is small because the correct feature quantities are selected and other marker data can compensate for an error. In addition, in this paper, we proposed a method to recognize the activity using the data when the human body template preparation required before Mocap data measurement is omitted, which is one of marker body matching work. The verification showed that even if the marker body matching operation was omitted, it was possible to recognize the action with high accuracy.

DOI： 10.1145/3341162.3345591

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85072894653&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Although activity recognition has been studied for a long time now, research and applications have focused on physical activity recognition. Even if many application domains require the recognition of more complex activities, research on such activities has attracted less attention. One reason for this gap is the lack of datasets to evaluate and compare different methods. To promote research in such scenarios, we organized the Open Lab Nursing Activity Recognition Challenge focusing on the recognition of complex activities related to the nursing domain. Nursing domain is one of the domains that can benefit enormously from activity recognition but has not been researched due to lack of datasets. The competition used the CARECOM Nurse Care Activity Dataset, featuring 7 activities performed by 8 subjects in a controlled environment with accelerometer sensors, motion capture and indoor location sensor. In this paper, we summarize the results of the competition.

DOI： 10.1145/3341162.3345577

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85072882944&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The recognition of complex and subtle human behaviors from wearable sensors will enable next-generation human-oriented computing in scenarios of high societal value (e.g., dementia care). This will require large-scale human activity corpuses and much improved methods to recognize activities and the context in which they occur. This workshop deals with the challenges of designing reproducible experimental setups, running large-scale dataset collection campaigns, designing activity and context recognition methods that are robust and adaptive, and evaluating systems in the real world. We wish to reflect on future methods, such as lifelong learning approaches that allow open-ended activity recognition. This year HASCA will welcome papers from participants to the Second Sussex-Huawei Locomotion and Transportation Recognition Competition and Open Lab Nursing Activity Recognition Challenge in special sessions.

DOI： 10.1145/3341162.3347765

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85072892160&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Human activity recognition is a potential area of research. For better performance, it requires significant amount of labelled data. Collecting labeled activity data is expensive and time-consuming. To solve this problem, transfer learning has been demonstrated very effective as it gathers knowledge from labeled train data of source domain and transfers that knowledge to target domain, which has little or no labeled data. In this paper, we propose unsupervised transfer learning from source dataset to target dataset, which are completely different in terms of number of users and samples. We have used Maximum Mean Discrepancy (MMD) based transfer learning model and compared with base Convolutional Neural Network (CNN) model. We have used 4 datasets for experiment. We have trained the model on a source dataset and then transferred the model to a target dataset, which has no labels to classify activities. We have found that transfer learning model has achieved better performance compared to the base model.

DOI： 10.1109/DASC/PiCom/CBDCom/CyberSciTech.2019.00155

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075116858&origin=inward

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：日本語   掲載種別：研究論文（研究会，シンポジウム資料等）

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The recognition of complex and subtle human behaviors from wearable sensors will enable next-generation human-oriented computing in scenarios of high societal value (e.g., dementia care). This will require large-scale human activity corpuses and much improved methods to recognize activities and the context in which they occur. This workshop deals with the challenges of designing reproducible experimental setups, running large-scale dataset collection campaigns, designing activity and context recognition methods that are robust and adaptive, and evaluating systems in the real world. We wish to reflect on future methods, such as lifelong learning approaches that allow open-ended activity recognition. Unique this year, HASCA will welcome papers from participants to the Sussex-Huawei Locomotion and Transportation Recognition Competition in a special session.

DOI： 10.1145/3267305.3274145

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85058343540&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this paper we summarize the contributions of participants to the Sussex-Huawei Transportation-Locomotion (SHL) Recognition Challenge organized at the HASCA Workshop of UbiComp 2018. The SHL challenge is a machine learning and data science competition, which aims to recognize eight transportation activities (Still, Walk, Run, Bike, Bus, Car, Train, Subway) from the inertial and pressure sensor data of a smartphone. We introduce the dataset used in the challenge and the protocol for the competition. We present a meta-analysis of the contributions from 19 submissions, their approaches, the software tools used, computational cost and the achieved results. Overall, two entries achieved F1 scores above 90%, eight with F1 scores between 80% and 90%, and nine between 50% and 80%.

DOI： 10.1145/3267305.3267519

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85058337765&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Telemedicine is an emerging challenge for the shortage of qualified professionals, particularly in under-resourced regions. Physical assessment by a non-medical doctor is a practice in telemedicine which discovers essential symptom of a patient who needs to consult a doctor. We aim at facilitating this stage with a conversational chatbot which identifies the patient by conversation. Adopting the procedures of physical assessment one critical types of conversation involves in the self-diagnosis. Further, it turned out that useful kinds of questions in chatbot at this stage are related to Yes/No questions. We discovered that particular difficulties lie in the ambiguous replies by the patients: a patient modifies a question which makes them answer yes or no, a response does not the corresponding reply to the question, a reply involves some part yes and some part no, and so on. Focusing on this particular type of question we introduce a text classifier using Long Short-Term Memory (LSTM) and build a corpus using Twitter.

DOI： 10.1145/3267305.3267704

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85058328754&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

We propose a new method for human activity recognition using a single accelerometer sensor and additional sensors for training. The performance of inertial sensors for complex activities drops considerably compared with simple activities due to inter-class similarities. In such cases deploying more sensors may improve the performance. But such strategy is often not feasible in reality due to costs or privacy concerns among others. In this context, we propose a new method to use additional sensors only in training phase. We introduce the idea of mapping the test data to a codebook created from the additional sensor information. Using the Berkeley MHAD dataset our preliminary results show this worked positively; improving in 10.0% both the average F1-score and the average accuracy. Notably, the improvement for the stand, sit and sit to stand activities was higher, typical activities for which the inertial sensor is less informative when using the wrist-worn accelerometer.

DOI： 10.1145/3267305.3267596

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85058340955&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

We propose a method to translate between multi-modalities using an RNN encoder-decoder model. Based on such a model allowing to translate between modalities, we built an activity recognition system. The idea of equivalence of modality was investigated by Banos et al. This paper replaces this with deep learning. We compare the performance of translation with/without clustering and sliding window. We show the preliminary performance of activity recognition attained the F1 score of 0.78.

DOI： 10.1145/3267305.3267512

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85058312644&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Motion capture generates data which are often more accurate than those captured by multiple of accelerometer sensors by their physical specification. Based on the observation that accelerometer data can be obtained by the second derivation of position data from motion capture, we propose a simulator, called MEASURed, for activity recognition classifiers. MEASURed can accommodate any number of virtual accelerometer sensors on the body based on some given motion capture data. Therefore, MEASURed can evaluate activity recognition classifiers in settings with different number, placement, and sampling rate of accelerometer sensors. Our results show that the F1-Score estimated by MEASURed is close to that obtained with the real accelerometer data.

DOI： 10.1145/3267305.3267509

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85058289678&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（研究会，シンポジウム資料等）

<p>IoT において，「歩く」，「立ち上がる」などの言語による行動のラベルを目的としたセンサからの行動認識を 可視化する技術は，「歩く」，「立ち上がる」という言語による行動のラベルを目的とした映像からの行動認識との 技術の融合を意味する．これは一転して,「歩く」，「立ち上がる」などの言語による行動のラベルのバイアスを排 除する新たな行動認識の形を提案し，新たなマイニングのモデルを提案する. ダンスの上手い人と下手な人のどこ が具体的に異なるかをセンサと映像からのマルチモダルな行動認識から探るプラットフォームの構築を報告する．</p>

DOI： 10.11517/jsaisigtwo.2018.AM-18_03

CiNii Article

CiNii Research

その他リンク： https://ci.nii.ac.jp/naid/130008080212

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper aims to detect the utterance which can be categorized as the breakdown of the dialogue flow. We propose a logistic regression-based and a Long Short-Term Memory (LSTM)-based methods. Using the input with utterance-response pairs, the performance of the LSTM-based method is superior to that of the logistic regression-based method in 36% measured with F-measure. We also measured the performance using the performance with utterance-response pairs: the performance with the input only with responses is unexpectedly inferior to those with responses in 6% to 23% measured with F-measure.

DOI： 10.1007/978-3-319-90740-6_18

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85046867198&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

DOI： 10.1515/pralin-2017-0036

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper introduces a new task, a recognition of multiple overlapping activities in the context of activity recognition. We propose a compositional CNN+LSTM algorithm. The experimental results show on the artificial dataset that it improved the accuracy from 27% to 43%.

DOI： 10.1145/3123024.3123095

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85030836480&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper describes the discourse parsing system developed at Dublin City University for participation in the CoNLL 2015 shared task. We participated in two tasks: a connective and argument identification task and a sense classification task. This paper focuses on the latter task and especially the sense classification for implicit connectives.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85072752049&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper describes our submission to the CoNLL-2015 shared task on discourse parsing. We factor the pipeline into sub-components which are then used to form the final sequential architecture. Focusing on achieving good performance when inferring explicit discourse relations, we apply maximum entropy and recurrent neural networks to different sub-tasks such as connective identification, argument extraction, and sense classification. The our final system achieves 16.51%, 12.73% and 11.15% overall F1 scores on the dev, WSJ and blind test sets, respectively.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84973379014&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（大学，研究機関等紀要）

arXiv

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper describes a new system combination strategy in Statistical Machine Translation. Tromble et al. (2008) introduced the evidence space into Minimum Bayes Risk decoding in order to quantify the relative performance within lattice or n-best output with regard to the 1-best output. In contrast, our approach is to enlarge the hypothesis space in order to incorporate the combinatorial nature of MBR decoding. In this setting, we perform experiments on three language pairs ES-EN, FR-EN and JP-EN. For ES-EN JRC-Acquis our approach shows 0.50 BLEU points absolute and 1.9% relative improvement obver the standard confusion network-based system combination without hypothesis expansion, and 2.16 BLEU points absolute and 9.2% relative improvement compared to the single best system. For JP-EN NTCIR-8 the improvement is 0.94 points absolute and 3.4% relative, and for FR-EN WMT09 0.30 points absolute and 1.3% relative compared to the single best system, respectively. © 2012 Springer-Verlag.

DOI： 10.1007/978-3-642-28601-8_4

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84858307159&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper presents two annotated corpora for word alignment between Japanese and English. We annotated on top of the IWSLT-2006 and the NTCIR-8 corpora. The IWSLT-2006 corpus is in the domain of travel conversation while the NTCIR-8 corpus is in the domain of patent. We annotated the first 500 sentence pairs from the IWSLT-2006 corpus and the first 100 sentence pairs from the NTCIR-8 corpus. After mentioned the annotation guideline, we present two evaluation algorithms how to use such hand-annotated corpora: although one is a well-known algorithm for word alignment researchers, one is novel which intends to evaluate a MAP-based word aligner of Okita et al. (2010b).

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85037351924&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper discusses Statistical Machine Translation when the target side is morphologically richer language. This paper intends to discuss the issues which are not covered by a factored translation model of Moses especially targetting EN-JP translation: the effect of Multi-Word Expressions, the separation of affixes, and other monolingual morphological issues. We intend to discuss these over a factored translation model. Copyright © 2011, Association for the Advancement of Artificial Intelligence. All rights reserved.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=80052409677&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper considers a scenario when we are given almost perfect knowledge about bilingual terminology in terms of a test corpus in Statistical Machine Translation (SMT). When the given terminology is part of a training corpus, one natural strategy in SMT is to use the trained translation model ignoring the given terminology. Then, two questions arises here. 1) Can a word aligner capture the given terminology? This is since even if the terminology is in a training corpus, it is often the case that a resulted translation model may not include these terminology. 2) Are probabilities in a translation model correctly calculated? In order to answer these questions, we did experiment introducing a Multi-Word Expression-sensitive (MWE-sensitive) word aligner and a hierarchical Pitman-Yor process-based translation model smoothing. Using 200k JP-EN NTCIR corpus, our experimental results show that if we introduce an MWE-sensitive word aligner and a new translation model smoothing, the overall improvement was 1.35 BLEU point absolute and 6.0% relative compared to the case we do not introduce these two. Copyright © 2011, Association for the Advancement of Artificial Intelligence. All rights reserved.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=80052406399&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（学術雑誌）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（学術雑誌）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（研究会，シンポジウム資料等）

Israel   Haifa   2011年01月23日  -  2011年01月27日

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The hierarchical Pitman-Yor process-based smoothing method applied to language model was proposed by Goldwater and by Teh; the performance of this smoothing method is shown comparable with the modified Kneser-Ney method in terms of perplexity. Although this method was presented four years ago, there has been no paper which reports that this language model indeed improves translation quality in the context of Machine Translation (MT). This is important for the MT community since an improvement in perplexity does not always lead to an improvement in BLEU score; for example, the success of word alignment measured by Alignment Error Rate (AER) does not often lead to an improvement in BLEU. This paper reports in the context of MT that an improvement in perplexity really leads to an improvement in BLEU score. It turned out that an application of the Hierarchical Pitman-Yor Language Model (HPYLM) requires a minor change in the conventional decoding process. Additionally to this, we propose a new Pitman-Yor process-based statistical smoothing method similar to the Good-Turing method although the performance of this is inferior to HPYLM. We conducted experiments; HPYLM improved by 1.03 BLEU points absolute and 6% relative for 50k EN-JP, which was statistically significant. © 2010 IEEE.

DOI： 10.1109/IALP.2010.34

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=79551520331&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（学術雑誌）

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（学術雑誌）

China   Beijing   2010年08月28日  -  2010年08月28日

This paper presents a new word align-
ment method which incorporates knowl-
edge about Bilingual Multi-Word Expres-
sions (BMWEs). Our method of word
alignment first extracts such BMWEs in
a bidirectional way for a given corpus and
then starts conventional word alignment,
considering the properties of BMWEs in
their grouping as well as their alignment
links. We give partial annotation of align-
ment links as prior knowledge to the word
alignment process; by replacing the max-
imum likelihood estimate in the M-step
of the IBM Models with the Maximum A
Posteriori (MAP) estimate, prior knowl-
edge about BMWEs is embedded in the
prior in this MAP estimate. In our exper-
iments, we saw an improvement of 0.77
Bleu points absolute in JP–EN. Except
for one case, our method gave better re-
sults than the method using only BMWEs
grouping. Even though this paper does
not directly address the issues in Cross-
Lingual Information Retrieval (CLIR), it
discusses an approach of direct relevance
to the field. This approach could be
viewed as the opposite of current trends
in CLIR on semantic space that incorpo-
rate a notion of order in the bag-of-words
model (e.g. co-occurences).

Kyutacar

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Parallel corpora are made by human beings. However, as an MT system is an aggregation of state-of-the-art NLP technologies without any intervention of human beings, it is unavoidable that quite a few sentence pairs are beyond its analysis and that will therefore not contribute to the system. Furthermore, they in turn may act against our objectives to make the overall performance worse. Possible unfavorable items are n : m mapping objects, such as paraphrases, non-literal translations, and multiword expressions. This paper presents a pre-processing method which detects such unfavorable items before supplying them to the word aligner under the assumption that their frequency is low, such as below 5 percent. We show an improvement of Bleu score from 28.0 to 31.4 in English-Spanish and from 16.9 to 22.1 in German-English. © 2009 ACL and AFNLP.

DOI： 10.3115/1667884.1667895

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=80052416271&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we give a description of the Machine Translation (MT) system developed at DCU that was used for our fourth participation in the evaluation campaign of the International Workshop on Spoken Language Translation (IWSLT 2009). Two techniques are deployed in our system in order to improve the translation quality in a low-resource scenario. The first technique is to use multiple segmentations in MT training and to utilise word lattices in decoding stage. The second technique is used to select the optimal training data that can be used to build MT systems. In this year's participation, we use three different prototype SMT systems, and the output from each system are combined using standard system combination method. Our system is the top system for Chinese-English CHALLENGE task in terms of BLEU score.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85133165029&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

France   Reims   2002年12月11日  -  2002年12月13日

Pervasive computing software help to manage information and reduce the complexity of available computing resources in a timely manner. On the other hand, a grid is a resource whose information is changing anytime and anywhere, where the availability of CPUs is only informed in run time when a cluster asks to the grid. On the other hand, a cluster is often implemented in a static way, which assumes some particular parallel architecture and the number of CPUs. Even when a cluster can consume maximum available CPU resources, if a cluster i implemented assuming the number of CPUs, a cluster could not run using more than this number of CPUs. Our mechanim of load balance protocol of cluster provides a dynamic way of implementing a cluster that can consume maximum available CPU resources on run time. In order to do so, we propoe two mechanisms: to provide yet another (graphical) parallel language to describe clusters and to provide the protocols between a cluster on a grid. While many paralell languages resolve parallel architecture dependencies in compile time, our parallel language resolve parallel architecture dependencies in run time. Our load balance protocol bases on this (graphical) parallel language and it provides implementation of them.

Kyutacar

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Estonia   Tallinn   2002年11月20日  -  2002年11月22日

This paper presents real-time formal language PRTccp, which complements priority-driven concerns in Tccp. First, we showed the necessity of formal language for priority-driven system compared to reactive real-time system. Secondly, we showed the grammar of PRTccp. Thirdly, we showed a small example of scheduler.

Kyutacar

開催期間： 2019年10月25日 - 2019年10月27日   記述言語：日本語  

開催期間： 2018年11月19日 - 2018年11月20日   記述言語：英語   開催地：Tel Aviv, Israel  

開催期間： 2018年09月13日 - 2018年09月14日   記述言語：英語  

開催期間： 2018年07月05日 - 2018年07月06日   記述言語：英語   開催地：Nagoya  

開催期間： 2014年10月22日 - 2014年10月26日   記述言語：英語   開催地：Vancouver, Canada  

デモ発表

科目区分：学部教養科目