A5036848200- Neural Networks and Applications
- Advanced Memory and Neural Computing
- Robotics and Automated Systems
- Neural Networks and Reservoir Computing
- Embedded Systems Design Techniques
- Advanced Neural Network Applications
- CCD and CMOS Imaging Sensors
- Robotics and Sensor-Based Localization
- Image Enhancement Techniques
- Neural dynamics and brain function
- Robotic Path Planning Algorithms
- Generative Adversarial Networks and Image Synthesis
- Image and Signal Denoising Methods
- Modular Robots and Swarm Intelligence
- Anomaly Detection Techniques and Applications
- Advanced Algorithms and Applications
- Color Science and Applications
- IoT-based Smart Home Systems
- Image Processing Techniques and Applications
- EEG and Brain-Computer Interfaces
- Fuzzy Logic and Control Systems
- Advanced Image and Video Retrieval Techniques
- Advanced Image Processing Techniques
- Nonlinear Dynamics and Pattern Formation
- Advanced Vision and Imaging
Kyushu Institute of Technology
2016-2025
Japan Society for the Promotion of Science
2021
Waseda University
2017
Tokyo University of Agriculture and Technology
2008-2012
The University of Tokyo
2010
Kyushu Institute of Information Sciences
2005
Abstract In contrast to AI hardware, neuromorphic hardware is based on neuroscience, wherein constructing both spiking neurons and their dense complex networks essential obtain intelligent abilities. However, the integration density of present devices much less than that human brains. this report, we molecular devices, composed a dynamic extremely network single-walled carbon nanotubes (SWNTs) complexed with polyoxometalate (POM). We show experimentally SWNT/POM generates spontaneous spikes...
Abstract A sulfonated polyaniline (SPAN) organic electrochemical network device (OEND) is fabricated using a simple drop‐casting method on multiple Au electrodes for use in reservoir computing (RC). The SPAN has humidity‐dependent electrical properties. Under high humidity, the OEND exhibits mainly ionic conduction, including charging of an electric double layer and diffusion. nonlinearity hysteresis current–voltage characteristics progressively increase with increasing humidity. rich...
Hardware‐based machine intelligence with the network architecture of reservoir computing (RC) is gaining interest because its biological computational resemblance along an easy and efficient neural training approach. Herein, such a physical RC (in‐materio RC) platform consisting recurrent formed by single‐walled carbon nanotube (SWNT)–porphyrin polyoxometalate (Por–POM) complex demonstrated. The executes fundamental properties nonlinearity, higher harmonic generation, 1/ f γ power law...
A time-domain analog weighted-sum calculation model is proposed based on an integrate-and-fire-type spiking neuron model. The applied to multi-layer feedforward networks, in which weighted summations with positive and negative weights are separately performed each layer summation results then fed into the next layers without their subtraction operation. We also propose very large-scale integrated (VLSI) circuits implement Unlike conventional voltage or current mode circuits, use transient...
The World Robot Challenge is an international competition for the social implementation of robots. Among them, Partner (Real Space) a category that focuses on domestic service robots, competing to achieve simple tasks such as tidying room, avoiding small obstacles, grasping specified object from shelf, and delivering waving person. In this category, we focused theme 'Keep Moving' worked researching developing technologies recognition, grasping, other tasks. For propose automatic dataset...
This paper proposes implementation of an Echo State Network (ESN) to Field Programmable Gate Array (FPGA).The proposed method is able reduce hardware resources by using fixed-point operation, quantization weights, which includes accumulate operations and efficient dataflow modules.The performance the designed circuit verified via experiments including prediction sine cosine waves.Experimental result shows that supports 200 MHz operation frequency facilitates faster computing ESN algorithm...
In this study, with the aim of installing an object recognition algorithm on hardware device a service robot, we propose Binarized Dual Stream VGG-16 (BDS-VGG16) network model to realize high-speed computations and low power consumption. The BDS-VGG16 has improved in terms accuracy by using not only RGB images but also depth images. It achieved 99.3% tests RGB-D Object Dataset. We have confirmed that proposed can be installed field-programmable gate array (FPGA). further Tiny, small XCZU9EG,...
Abstract Modern applications of artificial intelligence (AI) are generally algorithmic in nature and implemented using either general-purpose or application-specific hardware systems that have high power requirements. In the present study, physical (in-materio) reservoir computing (RC) was explored as an alternative to software-based AI. The device, made up a random, highly interconnected network nonlinear Ag 2 Se nanojunctions, demonstrated requisite characteristics in-materio reservoir,...
We propose a new fast learning algorithm for SOM and its digital hardware design based on the massively parallel architecture. When this proposed is realized by using Xilinx XC2V6000-6 FPGA, maximum performance of 17500 MCUPS achieved up to 256 competing units (16 /spl times/ 16 map) can be implemented. Each unit have weight vector which represented 128 elements bits accuracy. Furthermore, we applied realtime image enlargement system. In case full color (24 bits) from QQVGA (160 120 pixel)...
Our team Hibikino-Musashi@Home was founded in 2010. It is based Kitakyushu Science and Research Park, Japan. Since 2010, we have participated the RoboCup@Home Japan open competition open-platform league every year. Currently, has 24 members from seven different laboratories Kyushu Institute of Technology. home-service robots are used as platforms for both education implementation our research outcomes. In this paper, introduce technologies that implemented robots.
Reservoir computing (RC) is a framework for constructing recurrent neural networks with simple training rule and sparsely randomly connected nonlinear units. The network (called reservoir) generates complex motion that can be used many tasks including time series generation prediction. We construct reservoir based on the dynamics of pseudo-billiard system produce in high-dimensional hypercube. In particular, we use chaotic Boltzmann machine (CBM) whose units exhibit behavior interact each...
This paper proposes a shared synapse architecture for autoencoders (AEs), and implements an AE with the proposed as digital circuit on field-programmable gate array (FPGA). In architecture, values of weights are between synapses input hidden layer, output layer. utilizes less limited resources FPGA than which does not share weights, reduces amount modules used by half. For to be implemented into various types AEs, it three kinds parameters; one change number layers' units, bit width internal...
This paper proposes a method for the semi-automatic generation of dataset deep neural networks to perform end-to-end object detection and classification from images, which is expected be applied domestic service robots. In proposed method, background image floor or furniture first captured. Subsequently, objects are captured various viewpoints. Then, images composited by system (software) generate virtual scenes encountered robot. At this point, annotation files, will used as teaching...
Boltzmann machines (BMs) are useful in various applications but limited by their requirement to generate random numbers. In contrast, chaotic (CBMs) neural networks that imitate the stochastic behavior of BMs with dynamics and deterministic behavior, without CBMs can potentially require fewer hardware resources than original algorithms due unnecessity number generators. this study, hardware-oriented a differential multiply-accumulation operation proposed overcome difficulties implementing on...
This paper proposes a time-domain analog calculations model based on pulse-width modulation (PWM) approach for neural network including weighted-sum or multiply-and-accumulate calculation and rectified-linear unit operation.We also propose very-large-scale integration (VLSI) circuits to implement the proposed model.Unlike conventional voltage current mode circuits, our use transient operation in charging/discharging processes capacitors through resistors.Since calculate multiple...
Home service robots have pick-and-place tasks to grasp and carry objects.When performing pick-andplace tasks, need process search for objects in the case of missing a target object.We propose approaches robot object from shelf method selecting empty spaces move off-target objects.The proposed is based on two recognition models select space plan motions.The be moved selected size position spaces.In experiments, planned actions place an near group similar objects.We used these RoboCup@Home...