| CHEN TIEN HAO | M, 2回目発表 | 数理情報学 | 池田 和司, 金谷 重彦, 久保 孝富, 日永田 智絵 |
|
Title: Omics-data analysis for subtying delirium
Abstract: Delirium is a neuropsychiatric syndrome which is characterized in higher cortical functions impairments and is often occurred in the elderly who have received surgical treatments. Omics data is information generated by studies ending with -omics: genomics, proteomics, phenomics, etc. Our collected omics data with 286 post-operative patients which includes age, BMI, the assessment of delirium symptoms, biomarkers,...etc. However, we simply start by using the interested time-course of biomarkers to the research. In our study, we aim to make the prediction of a specific patient should develop delirium, and his/her worse severity of delirium by employing Generalized Estimating Equation to address the longitudinal data. In addition, we would like to understand whether the characterization of 7-phenotype groups based on the previous study can be achieved by the same biomarkers. Language of the presentation: English | |||
| 人見 謙太郎 | D, 中間発表 | 数理情報学 | 池田 和司, 松原 崇充, 久保 孝富, 日永田 智絵 |
|
title: Modeling of the human motor learning process aided by a training assist system
abstract: We aim to model the learning process of specific physical skills and thereby contribute to providing the design policy for an assist system that enhances learning. It was reported in a previous study that the variability of a trainee's motion is highly correlated to the trainee's quickness of learning. However, the large variability of the motion also indicates the instability of the trainee's physical control so we consider the variability solely is improbable for enhancing learning. By analyzing the dart-throwing training data of the previous study that proposed a training assist system that softly constrains the trainees' motion, we found a decrease in variability when using the assist system whereas it enhanced the learning. In this research, we apply Prioritized Exploration method that suppresses the variance of the sampling distribution, proposed in the context of the Black-Box Optimization, as a model to explain these effects of motion variability on the learning. language of the presentation: Japanese 発表題目: 訓練アシストシステムを用いた身体動作スキルの学習過程のモデル化 発表概要: 本研究では身体動作のスキルの学習過程のモデル化を試み、学習を高速化するアシストシステムの設計指針に結びつけることを目指す。 先行研究では訓練者の動作の Variability （変動性、ばらつき）の大きさが、スキル向上を高速化する重要な因子であると指摘されている。 しかし Variability は身体制御の不安定さでもあるため、単独でスキル向上を左右するとは考えにくい。 実際にダーツの訓練を訓練者の動作を緩やかに制限する訓練アシストシステムを提案した先行研究のデータを分析すると、ダーツのスキルは向上していたにもかかあわらず訓練中のVariabilityは減少していることが確認された。本研究では、ブラックボックス最適化の文脈で提案された探索分布の分散を抑制することで最適化を促進する Prioritized Exploration の拡張によってこの現象をモデル化する。 | |||
| 尾﨑 翔太 | M, 2回目発表 | 光メディアインタフェース | 向川 康博, 松原 崇充, 舩冨 卓哉, 藤村 友貴, 北野 和哉 |
|
title: Depth Estimation with Deep Depth-from-Defocus
abstract: The goal is to improve the accuracy of the Depth-from-Defocus (DFD) method for images captured by monocular cameras. For monocular images, a combination of deep learning and DFD has been used to achieve highly accurate depth estimation. However, prior methods have a problem that accuracy is poor for images taken with a different camera than the dataset used for training. Therefore, in this study, I propose a method that enables highly versatile depth estimation by deep learning that takes camera parameters into account, and present comparative results with previous studies. language of the presentation: Japanese 発表題目: 深層Depth-from-Defocusを用いた奥行推定 発表概要: 単眼カメラで撮影された画像に対し、Depth-from-Defocus(DFD)という手法の精度の向上を目的としている。単眼の画像では深層学習とDFDを組み合わせることで高精度の奥行推定を実現している。しかし、先行手法ではトレーニングに用いたデータセットと異なるカメラで撮影された画像では精度が悪いという問題点がある。そこで、本研究ではカメラパラメータを考慮した深層学習を行うことにより、汎用性の高い奥行推定を可能とする手法を提案し、先行研究との比較結果を示す。 | |||
| CHEN XIAN | M, 2回目発表 | 光メディアインタフェース | 向川 康博, 松原 崇充, 舩冨 卓哉, 藤村 友貴, 北野 和哉 |
|
Title: Predicting the shape of the organoid with generative model Abstract: The organoid is a kind of tiny, self-organized three-dimensional tissue cultures that are derived from stem cells. In the time of forming, the shape will change. If we can predict the shape of the organoid, we can find the organoid which can offer us the special shape of the organoid we want. There is few research based on the machine learning model to generate the shape of the organoid. In the research, I use the generation model based on the diffusion model to generate the new image of the organoid. We use the first several days organoid’s image captured by the microscope as input and the condition to control the generation of the next few days organoid’s shape and which part will change. In this research we can predict the change of the shape of the organoid without using the physical model of the organoid. The diffusion model shows the ability to predict the shape of the organoid in the next several days. Language of the presentation: English | |||