Biomedical Engineering Course Assoc. prof. Hiroshi DOUZONO E-mail:douzono [at] cc.saga-u.ac.jp (Please replace [at] with @ in e-mail address.) Academic Staff Database

Education: 1984 Kyoto Univ, Faculty of Engineering, Graduated 1986 Kyoto Univ, Graduate School, Division of Engineering, Master Course, Completed 1989 Kyoto Univ, Graduate School, Division of Engineering, Doctor Course, Accomplished credits for doctoral program Employment Experience: 1994-2007 Saga Univ., Faculty of Science and Engineering, Associate Professor 2007-2010 Saga Univ., Graduate School of Science and Engineering, Advanced Systems Control Engineering, Associate Professor 2010- Saga Univ., Graduate School of Science and Engineering, Advanced Technology Fusion, Associate Professor

Research Field : Soft computing, Bioinfomatics, Biometrics, Network Security

---

From the theory to the application of Self Organizing Map(SOM)
SOM is a neural network proposed by T.Kohonen. SOM can visualize the relation among high dimensional data on low dimensional space (generally 2 dimensional plane) with unsupervised learning. The research of SOM is proceeding in many universities and companies, and SOM is applied to many fields of application. Workshops of Self organizing map (WSOM) are held in every two years from 1997, and I attended and presented my research in 7 times.

Theory of Pareto-learning SOM

Pareto learning SOM is the SOM which uses novel learning algorithm proposed by H.Dozono. Conventional SOM uses the vector as input data. On the other hand, Pareto learning SOM used multi-modal vector which is composed of multiple vectors, and the learning algorithm is based on the Pareto optimizing theory. Pareto learning is superior to detect the anormal data which is not learned as training data. Pareto learning SOM is applied to biometric authentication and the analysis of packet traffic.

Hidden Markov Model Self Organizing Map (HMM-SOM)

Conventional SOM learns the input vector as reference vectors. HMM-SOM extract the models which generate the input vectors, and arrange them on the 2 dimensional plane according to the similarities among them. HMM-SOM is applied to the analysis of DNA sequences and the analysis of stock market.

Analysis of DNA sequence using Self Organizing Map

Recently, huge number of DNA sequences are generated by Next Generation Sequencer(NGS)s, and the development of the effective method which can handle them is required. As a such method, the method which uses SOM is under development. Specifically, SOM and Pareto-learning SOM which uses the frequencies of strings in the sequences and correlation coefficients of sequences are used as input vectors to SOM to classify the sequences.


Red:Human, Green:Mouse, Blue:Dog, Cyan:Paddy, Yellow:fly, Pink:Yeast

Visualization of the status of the students in C learning Class using SOM

It is important to identify the status of the students during the class which uses the computer for exercise. The system which visualize the status of students on 2 dimensional plane using the captured data of keyboard and mouse input of the students.

Development of the biometric authentication system using touch panel

Recently, network can be used everywhere with smart phones and tablet devices. The authentication system using touch panels equipped to these devices are developed, and SOM and Pareto-learning SOM are applied to the analysis and authentication algorithm.