Masato Odagaki

Variation Autoencoder Based Network Representation Learning for Classification.

Network representation is the basis of many applications and of extensive interest in various fields, such as information retrieval, social network analysis, and recommendation systems. Most previous methods for network representation only consider the incomplete aspects of a problem, including link structure, node information, and partial integration. The present study introduces a deep network representation model that seamlessly integrates the text information and structure of a network. The model captures highly non-linear relationships between nodes and complex features of a network by exploiting the variational autoencoder (VAE), which is a deep unsupervised generation algorithm. The representation learned with a paragraph vector model is merged with that learned with the VAE to obtain the network representation, which preserves both structure and text information. Comprehensive experiments is conducted on benchmark datasets and find that the introduced model performs better than state-of-the-art techniques.

Automatic motion tracking and data analysis system for a rat

The behavior analysis of animal has a key role in biology and robotics. In this paper, we choose the rat as analysis object. For understanding and grasping the behavior of rat, we set up an automatic motion tracking and data analysis system for a rat. This system includes three subsystems: inertial data acquisition, image tracking and real-time data analysis. The inertial subsystem uses one CC2530 development board and one small modular development board which integrates MPU6050 and electrical stimulator, and the inertial data collected by MPU6050 and the controls parameter sent by experimenter communicate by Zig-Bee wireless transmission module to realize the two-way transmission. The image tracking subsystem recognizes and tracks the rat by the camera. Besides, the inertial data and image tracking are time-synchronized. Finally, we program a data analysis interface by Matlab to conduct real-time data analysis. According to the tests, the automatic motion tracking and data analysis system can recognize the rat with a successful rate of 97%, combined with the inertial data. At the same time, we can conduct the real-time data analysis to simulate the attitude of the rat in the interface.

Deep Modeling of the Evolution of User Preferences and Item Attributes in Dynamic Social Networks

Modeling the evolution of user preferences and item attributes in a dynamic social network is important because it is the basis for many applications, including recommendation systems and user behavior analysis. This study introduces a comprehensive general neural framework with several optimal strategies to jointly model the evolution of user preferences and item attributes in dynamic social networks. Preliminary experimental results conducted on real-world datasets demonstrate that our model performs better than the state-of-the-art methods.

P3-2-05. Transcranial magnetic stimulation over the cerebellum facilitates excitability of spinal reflex in spinocerebellar ataxia

We reported that cerebellar transcranial magnetic stimulation (TMS) facilitates spinal reflex in healthy humans. The aim of this study was to investigate whether the “cerebellar” spinal facilitation (CSpF) appears in patients with spinocerebellar ataxia (SCA) with atrophy in the cerebellar cortex and dentate nucleus (DN). Two patients with SCA (one was SCA6 and the other was SCA31), participated in this study. Cerebellar inhibition (CBI) was tested using paired-TMS-paradigm with interstimulus intervals (ISI) of 1–8 ms in the right first dorsal interosseous muscle. To measure CSpF, we delivered TMS over the right cerebellum 100–130 ms before right tibial nerve stimulation, and CSpF was calculated with obtained conditioned/unconditioned H-reflex amplitude in the right soleus muscle. Voxel-based morphometry was used to verify the atrophy in cerebellar cortex and the DN. The results showed absence of CBI, but there was a significant facilitation of the H-reflex which occurred in the 120 ms ISI in both patients. These findings indicate that the pathways associated with the induction of CSpF and CBI are different, and that the cerebellar cortex and the DN are not required for inducing CSpF. Possible generator of CSpF could be other deep cerebellar nuclei or the brain stem.

Effective Strategies on Representing Information Networks

Network representation is the basis of many applications and of extensive interest in various fields such as information retrieval, social network analysis, and recommendation systems. Majority of previous methods on network representation only considered incomplete aspects of the problem, such as link structure, node information, or partial integration. The present paper proposes a comprehensive network representation model, which seamlessly integrates the text information, node label, and first-order and second-order proximity of a network. The effectiveness of the introduced strategies is experimentally evaluated. Results demonstrate that our method is better than state-of-the-art techniques.

A novel upper limb training system based on UR5 using sEMG and IMU sensors

This paper intends to design a system which acquires the trainers motion and force information in order to manipulate a robot arm applied for rehabilitations. Patients who suffering physical disability also can receive the professorial guiding and cheirapsis even excellent trainers are very busy and insufficient. The key point of this article is data acquisition and reconstruction of the movement of the upper limb by controlling the UR5 robot arm. Upper limbs postures are sensed by Inertial Measurement Unit (IMU) and transferred to STM32 microcontroller using I2C communication protocol. We employed the STM32 microcontroller to calculate attitude angles of both the upper arm and the forearm. And the method with using quaternions to calculate attitude angles is detailedly expounded in this paper. Besides, we employed the MYO armband to acquire upper limbs surface electromyography (sEMG) signals for estimating the muscle force of the upper limb. To verify the feasibility of the proposed system, we make three experiments including analyzing fluctuation range of the attitude angles from IMU signals, classifying muscle force using sEMG signals, and evaluating the effect of motion reconstruction. And the results show that the fluctuation range of acquired data are less than 1 degree, 4 typical motions of upper limb can be reconstructed. The proposed system can be used to reconstruct some upper limbs movement.

Touch interface for sensing fingertip force in mobile device using electromyogram

The aim of this study is to develop a three-dimensional touch interface for mobile devices, specifically a touch interface for detecting fingertip force. This interface consists of a conventional touch interface and an electromyogram (EMG) amplifier. The fingertip force during manipulation of the touch interface is estimated from the EMG measurement. We develop a method for obtaining fingertip force information using an EMG, while the two-dimensional position of the finger is measured using the conventional touch interface found in mobile devices. Further, we evaluate the validity of our newly developed interface by comparing the fingertip force estimated using our proposed method with the fingertip force measured using a force sensor. Lastly, we develop an application using our interface.

Optimization of Eddy Current Distribution Using Magnetic Substance in TMS

Transcranial magnetic stimulation (TMS) is a non-invasive method for stimulating the cortical neurons in the brain. The eddy current distribution is induced by the time-varying magnetic field of the magnetic coil used in the TMS. The brain neurons can be excited by the eddy current stimulation. The figure-of-eight coil in the TMS is used for the focal stimulation. The brain site beneath the central point of the figure-of-eight coil is conventionally determined as the stimulating site. However, it is difficult to find the optimal position to be stimulated by changing the coil. In this study, we propose an optimization method of the current distribution using a magnetic substance, without varying the coils position. We verify our method by a computer simulation.