Makoto Koshino

Inference of S-system models of genetic networks using Product Unit Neural Networks

In this study, we proposed the method of inference of genetic networks which expresses the regulation of genes. The proposed method does not solve the differential equations, learns the genetic networks using product-unit-neural-network (PUNN) and infer the S-system model of genetic networks which describes a set of differential equations. The experimental results show the proposal method is 160 times faster than the previous method which estimated S-system model of genetic networks while maintaining equivalent performance to the previous method.

An English vocabulary learning support system for the learner's sustainable motivation

In English vocabulary learning, continuation is an important factor; however, many learners are not good at continuing learning because they tend to prefer amusement or rest. Our proposed system is targeting learners who are eager to learn but are not able to continue learning for various reasons. We especially focused on English vocabulary learning, and described an approach for learners who have difficulty with continuing learning. Our developed application aggressively supports the learners’ sustainable motivation by gamification techniques and an efficient difficulty setting method.

Applying a path planner based on RRT to cooperative multirobot box-pushing

Considering robot systems in the real world, a multirobot system where multiple robots work simultaneously without colliding with each other is more practical than a single-robot system where only one robot works. Therefore, solving the path-planning problem in a multirobot system is very important.In this study, we developed a path-planner based on the rapidly exploring random tree (RRT), which is a data structure and algorithm designed for efficiently searching for multirobot box-pushing, and made experiments in real environments. A path planner must construct a plan which avoids the robot colliding with obstacles or with other robots. Moreover, in some cases, a robot must collaborate with other robots to transport the box without colliding with any obstacles. Our proposed path planner constructs a box-transportation plan and the path plans of each robot bearing the above considerations in mind.Experimental results showed that our proposed planner can construct a multirobot box-pushing plan without colliding with obstacles, and that the robots can execute tasks according to the plan in real environments. We also checked that multiple robots can perform problem tasks when only one robot could not transport the box to the goal.

Determining a smartphone's placement by material detection using harmonics produced in sound echoes

In this study, we propose a system to determine the placement of a smartphone using acoustic properties of the surface materials nearby. Detecting the surrounding materials allows the smartphone to change notification method automatically based on situational factors. Researchers have studied how to recognize the position in which smartphones are worn while walking, using accelerometer data; however it is difficult to identify a smartphones position while stationary, because the accelerometer value does not change significantly when the smartphone is put down. In this study, we developed a method to recognize surface materials close to a smartphone using echoes, based on an assumption that echoes of a selected frequency will differ in their properties, depending on the smartphones placement and the surface materials nearby. As a result of our experiment, our proposed method could classify six types of placement with 89.2% accuracy. We also revealed the limited influence of other environmental factors and the orientation of the smartphone, suggesting a rather robust result.

An experimental result of estimating an application volume by machine learning techniques

In this study, we improved the usability of smartphones by automating a user’s operations. We developed an intelligent system using machine learning techniques that periodically detects a user’s context on a smartphone. We selected the Android operating system because it has the largest market share and highest flexibility of its development environment. In this paper, we describe an application that automatically adjusts application volume.Adjusting the volume can be easily forgotten because users need to push the volume buttons to alter the volume depending on the given situation. Therefore, we developed an application that automatically adjusts the volume based on learned user settings. Application volume can be set differently from ringtone volume on Android devices, and these volume settings are associated with each specific application including games. Our application records a user’s location, the volume setting, the foreground application name and other such attributes as learning data, thereby estimating whether the volume should be adjusted using machine learning techniques via Weka.

Determining Smartphone’s Placement Through Material Detection, Using Multiple Features Produced in Sound Echoes

This paper proposes a system to determine the placement of a smartphone by using the acoustic properties of the surface materials nearby. Detecting the surrounding materials allows the smartphone to change its notification method automatically, based on situational factors. Researchers have studied how to recognize the position in which smartphones are worn while walking, using accelerometer data; however, it is difficult to identify a smartphone’s position while stationary, because the accelerometer value does not change significantly when the smartphone is put down. In this paper, we developed a method to recognize surface materials close to a smartphone, using echoes; this method is based on the assumption that echoes of a selected frequency will differ in their properties, depending on the smartphone’s placement and the surface materials nearby. Through our experiment, we found that our proposed method can classify 12 kinds of placement with 82.1% accuracy.

State magic: state estimation for Android smartphone

In this study, we propose State Magic, which estimates the smartphone state using various sensors as standard equipment. Smartphone state means smartphone location such as placing it in a pocket or putting it on a desk. If a smartphone can estimate its own state, developers can create various consumer support applications, such as an application for preventing mis-operations while the smartphone is stored in the pocket. Six test participants operated some action tasks to measure sensor values for our experiments. The experimental results show that our method can classify six situations with an 87.0% accuracy. The results also show two findings: (1) Using various sensors improves the estimation accuracy and (2) our method can classify smartphone states independently of the users action while the user is moving.

Constraint propagation + Ant Colony Optimization for automated school timetabling

This paper shows a hybrid ant algorithm for automated school timetabling, Really Full Look-ahead + Ant Colony Optimization (RFL+ACO). Previously, a constraint propagation based timetabling algorithm, Really Full Look-ahead Greedy (RFLG), has been proposed and has shown good results for some real timetabling problems. We adopt Ant Colony Optimization (ACO) to this algorithm in order to iteratively learn a selection order of variables to be instantiated and a selection policy of values to be assigned. A performance evaluation experiment using a real timetable data of unified lower and upper secondary school has been conducted, and results show that the proposed algorithm can construct good timetables.