Authorship：Lead author   Language：Japanese  

There is growing interest in time-series anomaly detection, which automatically determines the conditions of infrastructure or manufacturing equipment such as normal or abnormal by collecting time-series data from sensors and analyzing it using AI. This paper describes key technical challenges in the infrastructure and manufacturing fields, including AI's explainability, anomaly data collection, risk management of AI misjudgments, and data reliability. It then introduces a shapelet method as an explainable diagnostic technique and presents the extended versions that address the remaining challenges, along with their applications in the industrial fields.

DOI： 10.20729/00218781

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

Shapelets are subsequences that are effective for classifying time-series instances. Joint learning of both classifiers and shapelets has recently been studied because this approach improves algorithmic complexity and classification performance. However, the existing methods lack the power of feature discrimination due to using traditional sigmoid cross-entropy loss functions. To enhance feature discriminability, we propose self-adaptive scaling of the loss functions, inspired by the recent discriminative loss in computer vision. In addition, we propose a theoretically sound regularization that enhances feature discriminability and maintains shapelet interpretability by shrinking appropriate features. Using UCR datasets, we demonstrate improved area under the curve and interpretability of shapelets with a small number of shapelets.

DOI： 10.1137/1.9781611977172.22

Scopus

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

Shapelets are subsequences that are effective for classifying time-series instances. In this study, we consider when each time-series instance is obtained as progress, and formulate the problem of learning shapelet evolution over progress. For example, shapelets can change their shapes according to progress with human habituation, seasonal effects, and system degradation. When given time-series instances, progress values, and binary class labels, the proposed optimization formulation can jointly learn not only the shapelets and a classifier but also regression models for predicting shapelet evolution. The derived optimization solution method allows regression models to be learned by using off-the-shelf regression solvers, and scales linearly with time-series length. We demonstrate its effectiveness in industrial case studies.

DOI： 10.1109/ICDE53745.2022.00064

Scopus

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

Shapelets are time-series segments effective for classifying time-series instances. Joint learning of both classifiers and shapelets has been studied in recent years because this approach provides both superior classification performance and interpretable results. However, the optimization formulation is nonconvex, so bad local minima must be avoided. Very recently, this issue has been tackled by introducing Self-Paced Learning (SPL) into these methods. With the aim of intelligently discovering initial shapelets, we introduce two steps into this binary classification method so as to consistently optimize the same loss function of interest: optimizing discovery of discriminative initial shapelets from many time-series segments by using supervised feature selection, and jointly optimizing shapelets, model parameters, and latent instance weights in SPL. Using UCR datasets, we demonstrate improved area under the curve, and interpretability of shapelets where the number of shapelets is small.

DOI： 10.1137/1.9781611976700.30

Scopus

Authorship：Lead author   Language：English   Publishing type：Research paper (scientific journal)  

Shapelets are discriminative segments used to classify time-series instances. Shapelet methods that jointly learn both classifiers and shapelets have been studied in recent years because such methods provide both interpretable results and superior accuracy. The partial area under the receiver operating characteristic curve (pAUC) for a low range of false-positive rates (FPR) is an important performance measure for practical cases in industries such as medicine, manufacturing, and maintenance. In this article, we propose a method that jointly learns both shapelets and a classifier for pAUC optimization in any FPR range, including the full AUC. In addition, we propose the following two extensions for shapelet methods: (1) reducing algorithmic complexity in time-series length to linear time and (2) explicitly determining the classes that shapelets tend to match. Comparing with state-of-the-art learning-based shapelet methods, we demonstrated the superiority of pAUC on UCR time-series data sets and its effectiveness in industrial case studies from medicine, manufacturing, and maintenance.

DOI： 10.1089/big.2020.0069

Publishing type：Research paper (international conference proceedings)  

Large amounts of time-series data have become accessible due to the rapid development of Internet-of-Things technologies and the demand for extracting useful knowledge from these data is increasing. Toward this goal, time-series segmentation — dividing data into similar segments — is a promising method for understanding the mechanisms in time-series data. In this paper, we focus on time-lag that appears in real datasets. Time lag — a typical phenomenon in time-series data — occurs when the speed of information diffusion differs between variables. However, conventional methods cannot distinguish differences in segmentation positions. In response, we propose Lag-Aware Multivariate Time-Series Segmentation (LAMTSS), an algorithm capturing time-lag across variables to determine segmentation positions for each variable. LAMTSS utilizes dynamic time warping without hyperparameter tuning. We confirm the accuracy of LAMTSS using artificial datasets and demonstrate the discovery of useful knowledge in real datasets.

DOI： 10.1137/1.9781611976236.70

Scopus

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

Shapelets are time-series segments effective for classifying time-series instances. Joint learning of both classifiers and shapelets has been studied in recent years, because such methods provide both interpretable results and superior accuracy. Partial Area Under the ROC curve (pAUC) for a low range of False Positive Rates (FPR) is an important performance measure for practical cases in industries such as medicine, manufacturing, and maintenance. In this study, we propose a method that jointly learns both shapelets and a classifier for pAUC optimization in any FPR range, including the full AUC. We demonstrate superiority of pAUC on UCR time-series datasets and its effectiveness in industrial case studies.

DOI： 10.1137/1.9781611976236.1

Scopus

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

Shapelets are time-series segments effective for classifying time-series instances. Joint learning of both classifiers and shapelets has been studied in recent years because such a method provides both superior classification performance and interpretable results. For robust learning, we introduce Self-Paced Learning (SPL) and adaptive robust losses into this method. The SPL method can assign latent instance weights by considering not only classification losses but also understandable shapelet discovery. Furthermore, the adaptive robustness introduced into feature vectors is jointly learned with shapelets, a classifier, and latent instance weights. We demonstrate the superiority of AUC and the validity of our approach on UCR time-series datasets.

DOI： 10.1007/978-3-030-67658-2_34

Scopus

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

Shapelets are time-series segments effective for classifying time-series datasets. In recent years, the discovery of shapelets by classifier learning has been studied. Methods for shapelet discovery have attracted great interest because they provide not only interpretable results but also superior classifi-cation performance. However, they do not consider imbalanced classifications between majority and minority classes, which may occur in actual applications (e.g., anomaly detection). Our aim is to learn shapelets and classifiers using only training data for the majority class without the minority class. We propose a method called one-class learning time-series shapelets (OCLTS). OCLTS efficiently and simultaneously optimizes both the shapelets and a non-linear classifier based on a one-class support vector machine by a stochastic sub-gradient descent algorithm. Experimental results show the method's effectiveness for interpretability and imbalanced binary classification.

DOI： 10.1109/BigData.2018.8622409

Scopus

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)  

There is a need for forecasting of short-range future values in data streams such as traffic flows, stock prices, and electricity consumption. However, concept drift in non-stationary data streams is an important problem. We propose an online prediction method called OPOSSAM for such data streams. OPOSSAM manages time-series segments in short-term memory and long-term memory, and forecasts future values by local regression based on the similarity of time-series segments. In particular, OPOSSAM keeps long-term memory consistent by reducing redundant samples with large prediction errors, and automatically adjusts the prediction model based on short-term memory from the prior model learned from the entire memory in order to deal with concept drift. Experimental results show accuracy superior to that of baseline methods on real-world datasets of traffic flow, stock prices, and electricity consumption.

DOI： 10.1109/BigData.2018.8622585

Scopus

Authorship：Lead author   Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.2197/ipsjjip.25.107

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.2197/ipsjjip.24.751

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/rtcsa.2015.31

Authorship：Lead author   Language：Japanese  

近年，自動車では自動運転や衝突回避などが研究開発され，複数のセンサに加えて車々間通信など入力データの到着タイミングやデータ量が変動する車外からのデータ活用が進められている．このようなセンサ情報処理では，センサからデータが発生してから処理が完了するまでのEnd-to-Endのデッドラインをミスしないリアルタイム制約が要求される．データストリーム処理では，低遅延なデータ処理を実現しながら，クエリにより複雑なセンサ情報処理の開発効率を高めることができ，これまで平均遅延時間の削減など様々な目的に応じたスケジューリング方式がネットワークや金融サービスなどの分野で多く研究されてきた．しかし，ストリーム処理で用いられる従来方式は，リアルタイム制約の維持を目的としておらず，この目的の達成には適切ではない．本論文では，リアルタイムスケジューリングのアルゴリズムであるEarliest Deadline Firstに基づくストリーム処理のスケジューリング方式を，これらのセンサ情報処理に適用可能な方式として実現する．これにより，データ量が増加しても優先度の高いデータ処理を遅らせずに処理できる．車々間通信からの入力データ量が増加する場合における車両衝突警告を想定して，提案方式を評価した．その結果，従来方式と比較して，デッドラインミスを削減し，リアルタイム制約を維持しながら車々間通信からの入力データを多く処理することで車両衝突事故の削減に有効であることを確認した．
Recent automotive systems use a variety of sensor data and communications from outside the vehicle to promote autonomous and safe driving. Such sensor data processing requires to maintain real-time constraints, which require to meet End-to-End deadlines between when data is read from a sensor and when it is processed. Data stream processing eases to design the complicated data processing by query description, and process data at a low latency. Scheduling of stream processing has well been studied according to various purposes such as reduction of average latency, in fields such as networks and financial services. However, the existing methods used in stream processing are not intended to maintain the real-time constraints, and are not appropriate to achieve this purpose. In this paper, we propose scheduling methods of stream processing, based on Earliest Deadline First, which is an algorithm of real-time scheduling, so that the method can be applied to the automotive sensor data processing. By the method, higher priority data can be processed without delaying when increasing the data volume. We evaluated the method by assuming the vehicle collision warning in case to increase the input volume of data from vehicle-to-vehicle (V2V) communications. As a result, we confirmed that the proposed method reduced the deadline miss and the vehicle crash by processing more input data from V2V communications while meeting the deadlines, comparing to the existing methods.

CiNii Books

CiNii Research

J-GLOBAL

Authorship：Lead author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

DOI： 10.1109/ICDE.2015.7113377

Other Link： https://dblp.uni-trier.de/db/conf/icde/icde2015.html#YamaguchiNSIWHT15

Publishing type：Research paper (scientific journal)  

DOI： 10.4236/jtts.2015.52010

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.14569/ijacsa.2015.060240

Publisher：IEEE  

In the automotive field, intelligent control systems and information services using vehicle data are rapidly expanding. However, there is no software architecture to efficiently handle sensor data in vehicles and their surroundings. In order to address these problems and develop automotive software with higher productivity and trustworthiness, we have researched and developed data-centric software architecture for a data-stream management system for an automotive embedded system (eDSMS) and a stream local dynamic map (LDM) using eDSMS. An LDM is a collection of data that is intended to be used in cooperative intelligent transportation systems. In this paper, we present features of eDSMS and LDM using eDSMS, and demonstrate collision warning application software with eDSMS and a driver assistance system with a stream-based LDM.

DOI： 10.1109/uic-atc-scalcom.2014.29

CiNii Research

Publisher：IEEE  

In the automotive field, intelligent control systems and information services using vehicle data are rapidly expanding. However, there is no software architecture to handle sensor data in vehicles and their surrounding efficiently. To solve these problems, we have researched and developed data centric software architecture for automotive systems. We have also developed a data-stream management system for automotive embedded system (eDSMS) as data centric software architecture. In this paper, we design eDSMS in the Android platform by extending the in-vehicle eDSMS to provide drivers and users with services on the basis of the embedded data processing of eDSMS in a vehicle. We create a class that manages stream processing in eDSMS in the Android platform and a communication method between Android applications and eDSMS in the Android platform with the stream. Application program interfaces for eDSMS in the Android platform and a sample program using the APIs are presented.

DOI： 10.1109/uic-atc.2013.39

CiNii Research

Authorship：Lead author  

Language：Japanese   Publisher：東京 : 電子情報通信学会情報・システムソサイエティ  

CiNii Books

CiNii Research

Other Link： https://ndlsearch.ndl.go.jp/books/R000000004-I024135235

Language：Japanese   Publisher：Forum on Information Technology  

CiNii Books

CiNii Research

J-GLOBAL

J-GLOBAL

Total pages：606p   Language：Japanese  

CiNii Books

DOI： 10.1515/9781501504129-003

Presentation type：Public lecture, seminar, tutorial, course, or other speech  

Presentation type：Oral presentation (invited, special)  

Presentation type：Public lecture, seminar, tutorial, course, or other speech  

Language：Japanese   Publisher：一般社団法人 人工知能学会  

DOI： 10.11517/jsaikbs.127.0_06

CiNii Research

Authorship：Lead author   Language：Japanese   Publisher：東京 : 技術情報協会  

CiNii Books

CiNii Research

Other Link： https://ndlsearch.ndl.go.jp/books/R000000004-I032524458

Authorship：Lead author  

Authorship：Lead author  

J-GLOBAL

Authorship：Lead author   Language：Japanese   Publisher：東京 : 日本工業出版  

CiNii Books

CiNii Research

Other Link： https://ndlsearch.ndl.go.jp/books/R000000004-I031307250

Language：Japanese   Publisher：東京 : 東芝技術企画部  

CiNii Books

CiNii Research

Other Link： https://ndlsearch.ndl.go.jp/books/R000000004-I032361790

Language：Japanese   Publisher：The Japanese Society for Artificial Intelligence  

Along with the development of IoT technology, large amount of time series data are becoming available in recent years. To discover useful knowledge from time series data, the method of segmenting multivariate time series data into characteristic patterns has been receiving much attention. However, positions of segmentation obtained by previous methods are identical across variables, which makes difficult to capture the specific feature of each variable. To deal with this problem, we propose a new method that can obtain appropriate positions of segmentation for each variable. Moreover, we experimentally show the effectiveness of our proposed method using both artificial and real datasets.

DOI： 10.11517/pjsai.jsai2019.0_1i4j205

CiNii Research

Authorship：Last author  

Authorship：Lead author   Language：Japanese   Publisher：東京 : 電気学会  

CiNii Books

CiNii Research

Other Link： https://ndlsearch.ndl.go.jp/books/R000000004-I025840590

Language：Japanese  

近年，自動車には多くの車載センサが搭載されており，センサから取得された運転データを用いて様々な安全運転支援システムが研究されている．これらの安全運転支援システムは，現在のセンサ情報をもとに，目の前にある危険を検知することを目的としている．新たなアプローチとして，運転行動をモデル化するという試みがなされている．これらのモデルを利用することで，リアルタイムに取得される運転データをもとに現在の状況を動的に判断することが可能となる．本研究では，車載センサを用いて取得された運転データをもとに確率モデルを作り，安全運転を支援するシステムの検討を行う．

CiNii Books

CiNii Research

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

Local Dynamic Map(LDM) has been studied for ITS Cooperative Systems. The LDM is a hierarchical system that manages driving environment recognition information sent from vehicle on-board sensors and through vehicle-to-vehicle communications or map data. We have proposed "Stream-LDM" using Data Stream Management System (DSMS) to realize high response. However, we found processing time for map data on data base to be performance bottlenecks. In this paper, we describe that this problem is resolved by Map Data Processing Mechanism which makes it possible to process static map data in DSMS. By Evaluation using vehicle driving simulation, we confirmed that the average performance of spatial operator for map data to be improved several 10 times as much as the conventional methods.

CiNii Books

CiNii Research

J-GLOBAL