モデリング・応用、その他 Modeling and applications: Others

P2-1-251 組み込み用神経回路モデルLimited General Regression Neural Networkを使ったDC-DCコンバータのモデルベース制御 Model-based control for DC-DC converter using Limited General Regression Neural network ○近藤勇祐1, 山内康一郎1 ○Yusuke Kondo1, Koichiro Yamauchi1 中部大学工学部　情報工学科1 Departmemtn of Information Science, Chubu University1 In this paper, we propose a model-based control system for a DC-DC converter using an incremental learning neural network. In these days, most DC-DC converters are designed to control its chopper circuits using a small micro computer. Therefore, we can construct a quick and intelligent controller by embedding sophisticated software into the micro computer. One approach to realize it is replacing the former feed-back controller with a model-based controller. Normally, the model based controller should be designed properly before hand according to a huge log-data that reflects the properties of the target device. Therefore, normally, we need a high cost for constructing the model-based controller. To overcome this problem, we tried to embed a limited general regression neural network(LGRNN), which was proposed by one of the author in 2011, into the microcomputer(H8-3069F). The LGRNN is able to continue incremental learning under a fixed memory capacity. The proposed controller consists of the normal PID controller and the LGRNN. The PID controller makes the LGRNN learn its outputs as the supervised signal. As a results, the LGRNN learns the properties of the target device gradually and yields correct control signal. After the learning, the LGRNN is used as the foward controller. The experimental results showed that the proposed controller controls the DC-DC converter quickly when sudden input voltage changes are occurred.

P2-1-252 マウスの行動パラメーターを司る神経回路の統計学的推定 Statistical inference of neural circuits responsible for behavioral parameters of mice ○高瀬堅吉1, 菊地賢一2, 小田哲子1, 黒田優1, 船戸弘正1,3 ○Kenkichi Takase1, Kenichi Kikuchi2, Satoko Oda1, Masaru Kuroda1, Hiromasa Funato1,3 東邦大・医・解剖・微細形態1, 東邦大・理・情報科学2, 筑波大・国際統合睡眠医科学研究機構3 Dept Anat, Toho Univ, Tokyo1, Dept Inform Sci, Toho Univ, Chiba2, IIIS, Univ of Tsukuba, Ibaraki3 One of key goals of neuroscience research is to identify neural circuits responsible for a variety of animal behaviors such as feeding, sleep/wakefulness, stress response, emotion, and cognition, and to elucidate the functional alterations of the neural circuits which cause neurological and psychiatric disorders. However, there have been a very limited number of research methods for visualizing neural circuits underlying the specific behavior. Here we propose a statistical method combined with c-Fos expression analysis, which enables us to infer functional neural circuits in behaving animals. A multivariate analysis of c-Fos expression in the whole brain of behaving animals provides the following information related to functional neural circuits. (i) Principal component analysis extracts a unitary functional neural circuit in the brain. (ii) Path analysis infers spatiotemporal activity pattern in the functional neural circuits. (iii) Correlation analysis between the principal component score and the performance of behavioral task suggests the functional neural circuit underlying the specific behavior. Moreover, the comparison of inferred neural circuits between wild-type and gene-modified mice allows the evaluation of the functional link from gene, neural circuits and behavior. Thus, the present approach will provide the useful information to elucidate the functional neural circuits underlying complex behaviors.

上部に戻る

前に戻る