Masaji Katagiri

Browser State Repository Service

We introduce browser state repository (BSR) service that allows a user to save and restore multiple independent snapshots of web sessions on a browser. At a later time, the user can retrieve any saved snapshot on a potentially different browser on a different device to continue any one of the chosen saved session in any order. The web session snapshot captures a complete browser running state, including the last page that appears on the browser, document object state, script state, values that a user enters in forms on the last page, browser history for back and forward pages, and cookies. BSR service consists of a browser plug-in that takes browser session snapshots, and a repository server that stores snapshots securely for each user. The main contribution of BSR service is that it decouples association between browser state and a device, in favor of association between browser state and its user.

GUI migration across heterogeneous Java profiles

Existing cross-platform graphical user interface (GUI) development tools do not support migrate-able GUIs as they do not consider any runtime concern, such as running state transformations. To address this problem, we introduce Scalable Graphical User Interface (SGUI). It allows GUI developers to construct a platform-independent GUI that can be migrated across heterogeneous Java profiles. In this paper, we will focus on two major problems in supporting migrate-able GUIs. First is layout and widget transformation, which describes how to layout a presentation after a GUI is migrated from one platform to another. Second is running state and event handling transformations, which describes how to transform running states and event handlings when a presentation is changed after a migration.

Efficient Anomaly Detection System for Mobile Handsets

A new anomaly detection system for mobile handsets has been proposed. In this system, software behavior that deviates from a model representing normal behavior is considered to be an anomaly. It is generally impossible to cover software behavior exhaustively by the model, which could adversely affect accuracy. In order to resolve this problem, the proposed system assesses the anomalousness of behavior not covered by the model. Moreover, this system needs to have a low overhead in order to be used in mobile handsets, which have less computational resource than a PC. The proposed system adopts an efficient feature for behavior assessment to achieve a high accuracy with a low overhead. This system is implemented on the ARM architecture, which is widely used in mobile handsets. Experimental results clarify that the performance overhead is reasonable and anomalous behavior can be detected accurately.

A Deep-Learning-Based Method of Estimating Water Intake

In Japan, which has become a very aged society, the increasing burden of nursing care is an issue. Services and systems related to automatic recording of healthcare management of elderly people have been proposed in order to reduce the burden of nursing care. Water intake is one of the items necessary for healthcare management of elderly people. However, it is not currently automated, which is a burden on caregivers. In the case of the conventional method, the swallowing sound is used for estimating the water intake. However, the estimation error for each subject is large. Accuracy of estimated water intake is improved by using deep learning. Specifically, three features, namely, mel frequency cepstral coefficient (MFCC), duration of water intake, and a RASTA filter auditory spectrum, are extracted from a subjects swallowing sound (which is thought to be highly correlated with water intake). A method of estimating water intake, which considers abstract features that are difficult for people to find, is proposed and verified. It is experimentally shown that RMSE of water intake estimated by the proposed method using deep learning is reduced compared with that estimated by conventional methods.

Family Structure Attribute Estimation Method for Product Recommendation System

This paper proposes a method that estimatesconsumer family-structure attributesby focusing on purchasing behavior. The method generatesa relevance model for each product type between family-structure attributes and purchasing histories beforehand based on a consumer panel survey. Random Forest, a machine learning method, is employed to generate the model. The proposed method facilitates provisioning of smart recommendations to the consumer family such as suggesting products that reflectthe family structure attributes of the consumer.

Estimating customer preference through store check-in histories and its use in visitor promotion

This paper proposes a method to estimate the preference of customers based on store check-in histories. The proposed method can distinguish the preferences of customers who have no purchase histories. We adopt a machine learning algorithm for model acquisition. The estimation results can improve the efficiency of visitor promotion campaigns and advertising campaigns. An actual visitor promotion trial indicates the effectiveness of the proposed method.