| 吉澤 知彦 | 1461011: D, 中間発表 | 池田 和司，金谷 重彦，久保 孝富 |
|
title: Physiological study of the striosome in the reward-based learning abstract: The striatum is a major cortical input site of basal ganglia. It consists of two compartments, which are striosome and matrix. The striosome has only the mu opiate receptor. And it is interesting that only striosomal medium spiny neurons have direct inhibitory projections to the dopaminergic neurons of substantia nigra pers compacta. Previous studies show that the dopaminergic neurons encode the reward prediction error in the reinforcement learning and the striatum represents action values. In this study, we hypothesize that the striosomal neural activities subtract expected reward from actual reward in computing reward predication error. To test this hypothesis, we will conduct striosome selective optical recording during head-fixed classical conditioning using transgenic mice. As the preliminary experiment, we trained wild type mice for odor classical conditioning and recorded striatal neural activities. Mice were able to learn odor-reward associations. In addition, some neurons responded to reward amount or odor stimulus. language of the presentation: Japanese 発表題目: 報酬ベース学習における線条体ストリオソームの生理学的研究 発 表概要: 大脳基底核の主要な入力部である線条体はμオピオイド受容体を持つストリオソームと，マトリックスと呼ばれる2つのコンパートメントからなり，興味深いこ とにストリオソームの直接路中型有棘ニューロンだけが黒質緻密部のドーパミン作動性ニューロンに抑制性に直接投射している．また，ドーパミン作動性ニュー ロンは強化学習における報酬予測誤差をコードし，線条体は行動価値をコードしていることが知られている． 本研究ではストリオソームの報酬予測誤差計算における機能が実際の報酬からの予測報酬の減算であるとの仮説を立て，その検証のためストリオソームニューロ ンが選択的にマーキングされた遺伝子組み換えマウスに古典的条件付けを行い，ストリオソームニューロンの活動を光学的に記録する．予備実験として野生型マ ウスを用いて古典的条件付けを頭部固定下にて匂いを条件刺激として用いて行った．その結果，匂いから報酬量を予測する行動が観察された．さらにその際の神 経活動を記録・解析したところ，予期される報酬量や匂いに反応するニューロンが観察された． | ||
| 秋澤 翔 | 1451002: M, 2回目発表 | 池田 和司，佐藤 嘉伸，久保 孝富 |
|
title: Development of a Prediction Method of Epileptic Seizure Using Electrocorticogram towards Seizure Suppression
abstract: Epilepsy is one of the most common neurological diseases, causes various outward effects such as uncontrolled jerky movement. One of the most serious aspects of epilepsy is that the seizure appears suddenly and unexpectedly to the patient. In order to reduce the risk of epilepsy and improve patients' quality of life, Seizure Prediction methods have been studied over the past 25 years. These prediction methods have many potential benefits for specific patients, however there still remains the cases which we cannot take these methods effectively. We will focus on each patients' specific condition and analyze thier intracranial-EEG data to improve the prediction methods. language of the presentation: Japanese | ||
| 田村 真一 | 1451074: M, 2回目発表 | 池田 和司，松本 裕治，山田 武士，久保 孝富 |
|
title: Nearest neighbor search method under variably-weighted multiple dissimilarities using a single graph index
abstract: This presentation introduces the nearest neighbor search method using a single index, which allows users to put weights on the multiple dissimilarities for each search trial. Recently, there is a large amount and variety of data online. The nearest neighbor search method with multiple dissimilarities is fundamental to find the desired item from them. To cope with the multiple dissimilarities with existing methods, we usually fix the weights beforehand or aggregate results afterward. However, these approaches do not work very accurately or very speedy. We propose the novel nearest neighbor search method which allows users to set combination weight of multiple dissimilarities at the search stage, by using a single graph index considering arbitrary weights. The experiment with the real image dataset shows that our index performs comparably to the conventional indices constructed with pre-fixed weights, although our index is constructed only once. language of the presentation: Japanese 発表題目: 単一のグラフ索引で複数の非類似度を調整できる最近傍探索手法 発表概要: 本発表では，単一の索引を用いながら，複数の非類似度の重みづけを探索のたびに調整できる最近傍探索手法を提案する． オンライン上に多種多様なデータがあふれている今日，その中から所望のアイテムを見つけ出すには複数の非類似度を取り扱える最近傍探索手法が求められている． しかしこれまでに提案されている手法で複数の非類似度を取り扱うには，事前に重みづけを固定するか，それぞれの非類似度での結果を統合する方法がとられることが多く，高速かつ高精度に探索をすることは難しかった． 本研究では，複数の非類似度に対する任意の結合重みを考慮した単一のグラフを索引に用いることで，探索時にはその結合重みを自由に設定できる最近傍探索手法を開発した． 実画像データセットを使った実験の結果，提案手法は索引を１回しか構築しないにもかかわらず，それぞれの重みで固定して作った従来の索引構造に匹敵する性能を達成することが確認された． | ||
| 清水 勇志 | 1451057: M, 2回目発表 | 杉本 謙二，笠原 正治，松原 崇充，南 裕樹 |
|
title: Autonomous distributed operation scheduling for PV power generation systems abstract: In this presentation, we focus on the autonomous distributed operation scheduling of photovoltaic (PV) power generation systems composed of multiple PV generators. The operation scheduling determines whether or not each PV generator supplies electric power under the three conditions: (i) the power supply and demand balance, (ii) the local generation for local consumption, and (iii) the reduction in frequency of ON/OFF switching. In this presentation, we propose a simple distributed controller. Then, we evaluate the usefulness of the proposed controller by numerical simulations. language of the presentation: Japanese | ||
| RODRIGUEZ RAMIREZ JUAN ESTEBAN | 1451126: M, 2回目発表 | 杉本 謙二，笠原 正治，松原 崇充，南 裕樹 |
|
title: CMA-ES based design for finite-level dynamic quantizers
abstract: Quantizers, which convert continuous-valued signals to discrete-valued ones, play a crucial role in the implementation of networked control systems. In this study, we focus on the design of finite level dynamic quantizers which are effective tools to overcome the performance degradation of the system due to quantization. The quantizer design is not an easy task since the design problem is non-linear and non-convex. In order to solve the quantizer design problem, this study adopts covariance matrix adaptation - evolutionary strategy (CMA-ES), which is a metaheuristic algorithm used for optimization. We show the effectiveness of CMA-ES for the design of the quantizer using numerical experiments and compare the performance of the algorithm with other metaheuristic based design methods. We reveal how CMA-ES gives the best performance. language of the presentation: English | ||