Increasingly, longitudinal behavioral data captured by various sensors are being analyzed to improve workplace performance. In this paper, we analyze the correlation between the regularity of workers' behavior and their levels of stress. We used a 23-month behavioral dataset for 18 workers that recorded their use of PCs and their locations in the office. We found that the principal eigenbehaviors extracted from the dataset with PCA represented typical work behaviors such as overwork using a PC and routine times for meetings. We found that more than 80% of each of the 18 workers' individual behaviors could be reconstructed using nine principal eigenbehaviors. In addition, the deviation ranges for the reconstruction accuracies were significantly different for workers in different positions. We conducted the correlation analysis between work behaviors of the workers and their stress level. Our results show a significant negative correlation (r > 0.69, p < 0.01) between the accuracy of reconstructed work behaviors and physical stress levels; and a significant positive correlation between the accuracy of reconstructed behavior and stress dissolution abilities. Our results suggest that the correlation between the stress level of workers and the regularity of their work behavior exists. This correlation will be useful for occupational healthcare.

This paper describes the design of multimodal robotic system, embodiment of multimodal interaction (voice, gestures, emotion, touch panel and dance) in assistive social robot (Matilda) for modeling group based and one-to-one interactions with technology adverse elderly in nursing homes. It describes the human-centered evaluation of Matilda based on quality, naturalness, user satisfaction and predictive accuracy based on first ever field trials in Australia. The multimodal social robots have facilitated breaking the technology barriers with the elderly leading to several aged care facilities and community centers showing interest in future trials.

This paper discusses the modeling, the observation, and the analysis of a social experiment, in which we built an online social network to observe interactions between people. We try to model the problem from the micro-economics aspect; we use the concept of utility to model people's behaviors in this experiment. Furthermore, we consider incentive reward as an external factor and observed how people's behavior would change based on how we give incentive reward. We assumed the scenario of content recommendation in an online social network and compared the two reward assignment rules: i) a user gets reward if she or he informs her/his friend of the pointer to the content and ii) a user gets reward if her/his friend she or he informed of the content pointer reacts to the content. We expect that, in the former, the social network is not activated because the information flow is unidirectional, while, in the latter, the social network can be activated because it stimulates people to react to the information they have received.

In this paper, we propose a ride-sharing system with harmonic aggregation of user demands to support both human satisfaction and resource efficiency, and report results of the simulation of our proposed system showing the efficiency of our method. Our proposed ride-sharing system alters user action time and harmonically aggregates user transfer by using flexible parts of users' demands toward their actions. We built a demonstration system visualizing our future vision in which the proposed ride-sharing method was represented as one scenario. In this system, the ride-sharing will be implemented as the joining of several small personal vehicles. We performed our simulations with basic scenarios including part of the proposed method and obtained results showing that the ride-sharing system with our proposed method has four-times higher efficiency than that without the method.

In this paper, we study probabilistic modeling of heterogeneously attributed multi-dimensional arrays. The model can manage the heterogeneity by employing an individual exponential-family distribution for each attribute of the tensor array. These entries are connected by latent variables and are shared information across the different attributes. Because a Bayesian inference for our model is intractable, we cast the EM algorithm approximated by using the Lap lace method and Gaussian process. This approximation enables us to derive a predictive distribution for missing values in a consistent manner. Simulation experiments show that our method outperforms other methods such as PARAFAC and Tucker decomposition in missing-values prediction for cross-national statistics and is also applicable to discover anomalies in heterogeneous office-logging data.

Recently, mobile object sharing systems including car-sharing systems have been gathering more attention as an ecological activity. In this paper, we propose a user-collaborative mobile object assignment system using location information of users. The optimal mobile object assignment is to minimize travel distances of mobile objects in the existing mobile object assignment system, and the system controls allocation of mobile objects. The system recommends the two users to hand over the mobile object to each other if it is the optimal way for users, for example, one user gets off the car and another user rides on the same car continuously. Our proposed system uses the location information of both current and next users to find the optimal way for both users, and informs them the information. Then, the system can trigger the independent collaboration of users by leaving a final decision up to them. Firstly our system calculates optimal mobile object assignment for both current and next users. Secondly, if it is efficient for both current and next users to change the destination of the current user, the system recommends the current user to change the current destination to the calculated one. Thereafter, the system gets an extra fee from the next user, and gives a part of that to the current user as incentive for change of destination when it is needed. Our system can achieve a more friendly system for both user and environment than the current system and solve the distribution problem of mobile objects. Also, we did multi-agent simulations with basic scenarios including a part of proposed method, and got results showing that the system was effective at increasing the number of people who were collaborate with each other.

Humans have always desired to guess the future in order to adapt their behavior and maximize chances of success. In this paper, we conduct exploratory analysis of future-related information on the web. We focus on the future-related information which is grounded in time, that is, the information on forthcoming events whose expected occurrence dates are already known. We collect data by crawling search engine index and analyze collective view of future time-referenced events discussed on the web.

Publish/subscribe represents a new paradigm for distributed content delivery. It provides an alternative to address-based communication due to its ability to decouple communication between the source and the destination. However, it has remained a challenge to devise a scalable overlay supporting expressive content-filtering while satisfying the desirable requirements large distributed systems should fulfill. Our goal is to build an efficient P2P publish/subscribe network where only interested nodes are involved in event dissemination, and the amount of overhead generated by network discovery and membership management is small. In order to do so, we use a Bloom filter based mapping scheme to map IDs to nodes' interests, in addition to a new interest proximity metric to forward events and to build nodes' routing tables. As for network discovery we propose a new approach we call ldquoshared interest approachrdquo. Our scheme ensures an upper bound of routing tables size that only depends on the size of the ID digest. To evaluate the algorithms proposed in this work we conducted simulations in both static and dynamic settings.

This paper describes a robotic agent that can promote communication when people first meet. When two people meet for the first time, a communication mediator can be important because people often feel stressed and cannot talk comfortably. Our agent reduces their stress by using embodied entrainment and thus promotes communication. In the research field of embodied entrainment, suitable timing of a nod or back-channel feedback has been discussed, but situations to communicate in are limited. We have developed an embodied entrainment control system that recognizes communication states and adapts to them accordingly. For this, we focus on effective non-verbal information for communication. Using this information, our agent helps a talker and a listener to alternate their roles appropriately. We conducted communication experiments with the agent and confirmed its effectiveness. In the experiments, we used and compared different representation of the agent: an embodied robot agent, a computer graphics agent, and no agent. We report the comparison results and discuss representations for communication agents.

In this paper, we will propose a novel technique for Future Opinion Extraction, a new task of Information Extraction. The system we built can extract automatically future opinions, building automatic queries for the Search API. We obtained an F-Measure of 81.8% for the filtering of future opinion candidates. Our method provides future opinions allowing us to identify future problems and future goals of our society in the forty next years. The total system, called ChronoSeeker, contains +20,000 future opinions.

We propose a new technology called the camera-typing interface, which can read printed characters such as URLs with a low-resolution camera. It realizes a type of ubiquitous information service using smart phones with cameras. This method includes two main advantages: automatic concatenation of sequential shots and automatic error correction by re-shooting. The automatic concatenation enables a user to take segmented images of a character string, thus a low-resolution camera can be used as the interface device. The automatic error correction enables the user to correct misrecognized characters just by retaking images around them, providing an easy and natural way of error correction. We present two experimental results to prove effectiveness of our method. Both results indicate that our proposed method is helpful.

In a handwritten word or sentence, deformation of each character pattern often depends on that of other character patterns due to their cursiveness or the writer's characteristic. In this paper, a new word recognition method is proposed, that takes into consideration the dependency of deformation between characters. The stochastic model used in our method, which is to say a bigram model of character patterns, is constructed on the assumption that there is a Markovian property underlying in the deformation of a character string, and has a learning algorithm based on the EM algorithm. Experimental results in ZIP code recognition show the effectiveness of our method.

We propose a maximum-likelihood approach to segmentation-based recognition of unconstrained handwriting text. The segmentation scores and recognition scores are transformed into posterior probabilities, and the likelihood function which is composed of both these probabilities and character n-gram probabilities is derived from the Bayesian theorem. The recognition result which maximizes the function can be obtained by Viterbi search. Experiments have shown that the proposed likelihood function is effective in the recognition of online Japanese text.

In this paper, we describe a new system for analyzing the motions involved in handling documents on a desk. Video images are captured by a camera mounted above the desk and the motion which occurs when someone handles the document is classified as movement of the document, turning of the pages, or finger pointing. Furthermore, the system can detect the appearance of fresh regions as a result of movement of the document and the turning of its pages and it can also detect the positions of fingers in the image to realize a natural pointing function. We evaluated implementations of some components of the system and found that they provide sufficient performance to act as parts of a user interface for an interactive document management system.

A new method for reconstructing a document from an image sequence and recognizing contents of the document is described. In this method frame images in the image sequence are acquired as partial sub-images by a moving camera. The reconstruction includes a rather expensive registration process. To make its use practical, the proposed method registers sub-images hierarchically by sub-tree matching of the layout structure. Images of a corresponding character pair are piled using sub-pixel precision displacement and interpolated to restore image quality and resolution. The proposed method increased the character recognition rate by 6.8 points to 93.1% when resolution was 133 dpi. The computational cost was reduced to 1/23 that of registration by conventional matching.

This paper reports a formulation of learning vector quantization (LVQ) using a new misclassification measure based on minimum classification error (MCE). We show that the convergence property of reference vectors depends on the definition of the misclassification measure, and show that our definition guarantees the convergence, unlike LVQ1.1 or Juan and Katagiri's formulation based on MCE (1992). Experimental results for handwritten digit recognition reveal that the proposed method is superior to LVQ algorithms in recognition capability.

Describes a method of retrieving handwritten text with handwritten queries. The properties of handwritten Japanese texts have prevented conventional handwritten European text retrieval methods from being applied to Japanese. To address this problem, the proposed method was devised to segment texts into blocks and search stored texts by using a two-stage flexible matching method. At the first stage, the features used for flexibly matching block series were designed to reduce the amount of the processing which produces features of combined blocks. In the second stage, flexible matching between corresponding blocks was employed to exclude erroneous candidates. Experiments demonstrated 99.5% recall and 20% precision in the case of queries with four characters.

We describe a new handwritten address recognition method which can correct the errors occurring in line extraction, character segmentation, and character recognition as a possible means of avoiding the error accumulation which occurs during the recognition sequence in conventional methods. We formulate the address recognition method as a minimum cost search problem. We define the character recognition cost which estimates the reliability of the character recognition result, the arrangement cost which estimates the plausibility of the character string's spatial arrangement, and the word knowledge cost which estimates the plausibility of the linguistic conditions. By using a combination of these costs, the proposed method can recognize an address which has not been extracted as a single line from input images by a conventional method. The efficiency of the proposed method is evaluated through an experiment using 600 Japanese mail images. An address recognition rate of 79.38% was obtained.

Describes a non-uniform sampling method for obtaining local orientation features to be used in the recognition of Japanese kanji characters, the proposed method determines feature extraction by sampling features one-dimensionally in a direction perpendicular to a given line orientation. Furthermore, it samples features at non-uniform intervals so as to avoid misidentifying two lines in close proximity as a single line. It has been tested with handwritten kanji characters, and the results show a good ability to distinguish such characters, a task which can easily confuse conventional recognition methods that rely on uniform sampling of local orientation features.

This paper describes a new method for analyzing Japanese mail layout. We first classify the many kinds of variations in Japanese mail layout into format variations, extraneous object variations, and arrangement variations of destination address characters. These variations are cyclically dependent on one another, but have insufficient information to fix other variations. To solve these problems, we developed an address layout analysis method using a hierarchical hypothesis verification process, which consists of a format determination process and a destination address character line detection process, hierarchically. This method increased the correct address format determination rate by 5.8% to 86.6% and the correct destination address character line detection rate by 10.2% to 92.8% for 1482 test pieces of mail. This method is practically applied to Japanese mail sorting machine.

This paper discusses sampling intervals of two-dimensional Gabor features in the two-dimensional pattern, orientation angle, and logarithmic frequency domains. The discussion on the feature stabilities for basic image transformations clarifies the stable range for translation, rotation and scaling. These stable ranges directly lead to sampling intervals for the individual feature domains. Multi-resolution features are constructed in terms of the combination of elements optimally sampled in the feature domains. The features of the optimal sampling intervals are examined in printed Japanese character recognition. The optimal sampling intervals maximize recognition rates at the same time as keeping computational cost low.

This paper describes the feature stabilities for four basic image transforma-tions, while examining an image feature extraction method using 2D Gabor filters, as an exmple. The basic transformations are intensity change, scaling, translation, and rotation. Based on the consideration about intensity change and scaling, feature normalization methods were proposed. Moreover, opti-mal sampling resolutions are determined according to the discussions. Scal-ing is related with the sampling resolution in the frequency domain and the filter wavelength should be y/2-fold of the smallest filter wavelength, where j is an integer number. The sampling interval in an image domain is a half of a filter wavelength. The sampling interval in a rotation angle domain is TT/10. The Gabor feature extraction method, whose sampling resolutions were determined based on the stability considerations, were examined on printed Japanese character recognition. When sufficient sampling resolu-tion was not attained, the recognition rate became lower. This method can achieve a maximally 96.8 % recognition rate, when every sampling resolution is satisfied. 1

It is shown that an offline character recognition method is effective for use in an online Japanese character recognition. Major conventional online recognition methods have restricted the number and the order of strokes. The offline method removes these restrictions, based on pattern matching of orientation feature patterns. It has been improved with developments in nonlinear shape normalization, nonlinear pattern matching, and the normalization-cooperated feature extraction method. It was used to examine 52,944 online Kanji characters in 1,064 categories. The recognition rate achieved 95.1%, and the cumulation recognition rate within the best five candidates was 99.3%.>

A feedback pattern recognition method based on an inverse recall neural network model is proposed. The model is a kind of multi-layer feedforward model. This model has three functional components. First is a recognition result generation according to a feedforward process in a multi-layer model. Another component calculates an uncertainty score from the output values. If the uncertainty score is greater than a threshold value, the last component, that is an inverse recall, operates. The inverse recall function produces changes in input values, in order to reduce the uncertainty score. The changes show input parts which are important for more certain recognition but are missed in the input pattern. The feedback method can adjust feature extraction parameters so as to detect the important features shown by the inverse recall network model. Then, features are extracted again by using the modified feature extraction parameter values. These feedforward and feedback processings are repeated until a certain recognition result is obtained. This method was examined for handwritten alpha-numeric recognition and it was found that a rejection ratio can be reduced at the same substitution error ratio.>

The authors propose a method to segment a character line image into individual character images and recognize them. To segment Japanese handwriting accurately, it is necessary to use character recognition results and contexts. However, recognition results might be wrong, or recognition confidence scores might be inaccurate. Dictionary consulting is not sufficient to deal with such ambiguous character recognition results. The paper reports on a method which hierarchically uses transitional information and a word dictionary for recognition results for all possible characters. Experimental results show that, for character line samples written roughly, 91.7% recognition rate is achieved while the recognition rate for a method without transitional information is 78.3%.>