Papers - YAMAMORI Kunihito
-
Hierarchical policy gradient method with the combination of reinforcement and auxiliary learning Reviewed
Masaya Yoshida, Kentaro Inoue, Masaru Aikawa, Kunihito Yamamori
Proceedings of the 24th International Symposium on Artificial Life and Robotics 706 - 710 2019.1
Language：English Publishing type：Research paper (international conference proceedings)
we aim to improve learning efficiency of reinforcement learning by hierarchical combination of Proximal
Policy Optimization (PPO) and auxiliary tasks. PPO is a method to stabilize learning. Auxiliary tasks are used in UNsupervised
REinforcement and Auxiliary Learning (UNREAL) which is a kind of policy gradient method based on Asynchronous Advantage
Actor-Critic (A3C) for high-speed learning convergence. We aim to improve learning efficiency by combining these two methods. We compare our method with previous methods using two benchmark tasks ‘Hopper’ and HumanoidStandup’ in ‘Gym’ provided by OpenAI. Experimental results showed that the proposed method achieved better performance than that by PPO. -
公開脆弱性データベースを利用した対応順位決定支援ツール
山森一人，齊藤燎，相川勝，井上健太郎
宮崎大学工学部紀要 47 305 - 309 2018.7
Language：Japanese Publishing type：Research paper (bulletin of university, research institution)
-
Enhanced Home Surveillance Robot System by using Raspberry Pi and Arduino
Win Su San, Kunihito Yamamori, Masayuki Mukunoki, Than New Aung
宮崎大学工学部紀要 47 325 - 329 2018.7
Language：English Publishing type：Research paper (bulletin of university, research institution)
Modern home security systems endorse monitoring and control functions in a remote fashion via devices such as laptops, PDAs, or cell phones1). This work represents a design and implementation of Wi-Fi enabled home surveillance robotic car which detects smoke, flame and intrusions discarding false positives. The robot's movement is controlled manually by the user through the developed GUI over the global internet. The communication between the robot and the user is established via the standard TCP/IP protocol and in case of emergency, the notification will be sent to the user via email. In any case, all sensory data (updated every 500 milliseconds) will be uploaded to the web site and available to the user all the time. Live feed from the camera will be viewable from any web browser.
-
A novel audio fingerprinting method based on music elements for similar music retrieval Reviewed
T.Aoshima, K.Yamamori, M.Aikawa, K.Inoue
Proc. 23rd International Symposium on Artificial Life and Robotics 452 - 456 2018.1
Language：English Publishing type：Research paper (international conference proceedings)
-
Lightweight virtual node protocol for large scale IoT device networks Reviewed
Y.Shinooka, K.Yamamori, M.Aikawa, K.Inoue
Proc. 23rd International Symposium on Artificial Life and Robotics 478 - 483 2018.1
Language：English Publishing type：Research paper (international conference proceedings)
-
Reinforcement learning approach for smart air traffic control support system Reviewed
M.Watanabe, K.Yamamori, M.Aikawa, K.Inoue
Proc. 23rd International Symposium on Artificial Life and Robotics 457 - 461 2018.1
Language：English Publishing type：Research paper (international conference proceedings)
-
Markov chain analysis of Leading Ones problem Reviewed
Du Y., Aoki K., Sakamoto M., Furutani H., Yamamori K.
Artificial Life and Robotics 22 ( 4 ) 443 - 448 2017.12
Language：English Publishing type：Research paper (scientific journal)
Evolutionary algorithms (EAs) are randomized optimization search techniques, and the theoretical study of the first hitting time is very important in the practical applications of EA. We investigate the first hitting time of Leading Ones (LO) problem theoretically on Random Local Search (RLS) algorithm and (1 + 1)EA by absorbing Markov chain. This paper reports the calculation method of the first hitting time on LO problem with (1 + 1)EA and RLS. We found that the results only depend on the length of string, and the hitting times can be given approximately by the quadratic function. In the numerical experiments, we showed the agreement between the results of the practical calculation and the theoretical results. Our results give the good example to predict the running time of the practical EA calculations by Markov chain method.
-
Three-step Recognition Method for Handwritten Digits Using SVM and Neural Network
Memoirs of Faculty of Engineering ( 46 ) 299 - 302 2017.7
Language：Japanese Publishing type：Research paper (bulletin of university, research institution)
-
Evaluation of Heterogeneous Crossover Island Model by TSP
Memoirs of Faculty of Engineering ( 46 ) 317 - 320 2017.7
Language：Japanese Publishing type：Research paper (bulletin of university, research institution)
-
Learning Efficiency Improvement of Reinforcement Learning in Restricted View Environment
Memoirs of Faculty of Engineering ( 46 ) 313 - 316 2017.7
Language：Japanese Publishing type：Research paper (bulletin of university, research institution)
-
Generalization Ability Improvement by Parallel Convolution and Dropout
Memoirs of Faculty of Engineering ( 46 ) 309 - 312 2017.7
Language：Japanese Publishing type：Research paper (bulletin of university, research institution)
-
Improving Software Development in Business Operation Effectively Approached by CMMI
Hnin Thandar Tun, T. Katayama, K. Yamamori, and Khine Khine Oo
宮崎大学工学部紀要 ( 46 ) 285 - 290 2017.7
Language：English Publishing type：Research paper (bulletin of university, research institution)
-
Early Detection of Lung Cancer by Artificial Neural Network and Fuzzy Inference System
Khaing Zin Htwe, K. Yamamori, T. Katayama, and Tin Mar Kyi
宮崎大学工学部紀要 ( 46 ) 321 - 326 2017.7
Language：English Publishing type：Research paper (bulletin of university, research institution)
-
犠打と犠飛を考慮したOERA改良モデルによる日本プロ野球打者の評価
山森一人，大塚蒼，相川勝
宮崎大学工学部紀要 ( 46 ) 303 - 307 2017.7
Language：Japanese Publishing type：Research paper (bulletin of university, research institution)
-
Markov chain analysis of Leading Ones Problem Reviewed
Yifei Du, Kenji Aoki, Makoto Sakamoto, Hiroshi Furutani, and Kunihito Yamamori
Proceedings of the 22nd International Symposium on Artificial Life and Robotics 520 - 524 2017.1
Language：English Publishing type：Research paper (international conference proceedings)
-
Development of Vulnerability Detection tool for Web Application based on MVC Framework Reviewed
Shoichiro Saruwatari, Kunihito Yamamori, Masaru Aikawa
Proceedings of the 22nd International Symposium on Artificial Life and Robotics 525 - 529 2017.1
Language：English Publishing type：Research paper (international conference proceedings)
-
Automated lung nodule classification by artificial neural network and fuzzy inference system
Htwe K., Yamamori K., Katayama T., Kyi T.
2016 IEEE 5th Global Conference on Consumer Electronics, GCCE 2016 2016.12
Language：Japanese Publishing type：Research paper (scientific journal) Publisher：2016 IEEE 5th Global Conference on Consumer Electronics, GCCE 2016
© 2016 IEEE. Lung cancer is the most cause of death from all cancers being less survival rate according to various online cancer societies. The former medical studies showed that the early diagnosis of lung cancer with screening program is the way to decline mortality rate. Radiological examination takes a tremendous time and need many medical experts having a large number of images per patient. Consequently, the accuracy of examination results mainly depends on the skills of physicians. Computer-aided Detection (CAD) in radiology can provide a functional and advantageous way to physicians aiming at improving accuracy and assisting in early detection of cancer, saving the time of radiologists in exam evaluation of cancers. Our proposed CAD system has fulfilled these all tasks combining image processing and Artificial Neuro-Fuzzy Inference System (ANFIS); with 98% sensitivity of finding true lung nodules, 94.4% sensitivity and 85% accuracy of removing false detectives, and 87% sensitivity and 78% accuracy for cancerous nodules detection.
-
Business goals monitoring and control measures in CMMI
Tun H., Katayama T., Yamamori K., Oo K.
2016 IEEE 5th Global Conference on Consumer Electronics, GCCE 2016 2016.12
Language：Japanese Publishing type：Research paper (scientific journal) Publisher：2016 IEEE 5th Global Conference on Consumer Electronics, GCCE 2016
© 2016 IEEE. Business goals monitoring and control measures are important process to achieve the target of business systematically in specified business area or project. Monitoring and control provide business goals with a measurable index to align directly in a specific area or project. In this paper, we generate all of the business process improvement goals by Capability Maturity Model Integration (CMMI) model. This procedure that is used to establish the business goals of an effective business area or related project work scope applied by Goal Question Metric (GQM) approach. This approach will be defined all of the business process goals and measurable index to perform business operation functionality. The metric program framework that is used in this paper to generate the metric plan is described, followed by a complete metric plan, which includes GQM Analysis to monitor and control measurements of defined goals systematically.
-
Asymmetric mutation model in genetic algorithm Reviewed
Yifei Du, Kenji Aoki, Makoto Sakamoto, Kunihito Yamamori, Hiroshi Furutani
Artificial Life and Robotics 2016.10
Language：English Publishing type：Research paper (scientific journal)
-
Automated Lung Nodule Classification by Artificial Neural Network and Fuzzy Inference System Reviewed
Khaing Zin Htwe, K.Yamamori, T.Katayama, Tin Mar Kyi
Proceedings of 2016 IEEE 5th Global Conference on Consumer Electronics 369 - 370 2016.10
Language：English Publishing type：Research paper (international conference proceedings)