| 大石 悠貴 | M, 2回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 加藤 博一, 内山 英昭, Perusquia Hernandez Monica, 平尾 悠太朗 |
|
発表題目: ARVR空間への適用に向けた三次元再構成手法の調査 発表概要: AR・VR分野において, フォトリアルな物体を掲示することは没入感を得るために必要な要素となります. しかし, 従来の古典的な3次元再構成手法では, 細かく複雑な部分や透明、不透明な物体に対して写実的な表現をすることが困難です. 私は, 高周波な細かい部分に対して高い表現力を持ち, 複雑な光の反射を考慮して学習を行うことが可能であり、鏡面や不透明な物体に対してもフォトリアリスティックな新視点画像の生成が可能なNeRF(Neural Radiance Field)に興味を持ち, この手法をテーマにした研究を行いたいと考え, 調査しています. Neural radiance fieldは約100枚～150枚程度の画像を学習することでニューラルネットワーク内で暗黙的に3次元再構成を行い、入力画像内にはない新視点画像の出力が可能な手法です. Title: Investigation of 3D reconstruction methods for application in ARVR space. Abstract: In the field of AR and VR, posting photorealistic objects is a necessary element for achieving an immersive experience. However, it is difficult to achieve photorealistic representation of fine and complex details and transparent and opaque objects using conventional classical 3D reconstruction methods. I am interested in NeRF (Neural Radiance Field), which has high expressive power for high-frequency details, can learn to account for complex light reflections, and can generate photorealistic new viewpoint images for specular surfaces and opaque objects. I would like to conduct research on this method and am investigating it. Neural radiance field is a method that can implicitly perform 3D reconstruction in a neural network by learning about 100 to 150 images, and can output a new viewpoint image that is not in the input image. | |||
| 末原 和樹 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸, 加藤 博一, 大竹 義人, SOUFI Mazen |
|
title: *** replace this part with the title of your talk IN ENGLISH ***
abstract: *** replace this part with the abstract of your talk IN ENGLISH *** language of the presentation: *** English or Japanese (choose one) *** 発表題目: *** 人工股関節患者の術前・術後の全身における ３次元形状と骨密度、筋肉量の推定 *** 発表概要: *** 筋骨格系の健康は持続可能な高齢化社会を確立する上で特に重要になっています。骨粗しょう症やサルコペニアのような筋骨格系疾患は、骨折を引き起こす運動器疾患の代表的なものであり、日常生活の品質や機能に重大な影響を与える可能性があります。そのため、筋骨格系の健康状態を頻繁に検査することは重要であり、少ない放射線量で正確に診断できる方法が期待されています。本研究では、直交する二方向から同時に撮影したx線画像を利用することで、骨密度と筋肉量の推定を目指す。 *** | |||
| 遠藤 力人 | M, 1回目発表 | 数理情報学 | 池田 和司, 加藤 博一, 久保 孝富, 日永田 智絵 |
|
title: Football players evaluation using reinforcement learning
abstract: Football clubs are keen to acquire and retain good players in order to achieve results. But there are many cases of overpriced players failing to perform well at all, and conversely, underpriced players playing very well. Therefore, we would like to use deep reinforcement learning to create a comprehensive football player evaluation model that properly evaluates how good a player is, regardless of his name recognition or current market price. language of the presentation: Japanese | |||
| 和田 有里菜 | M, 1回目発表 | 数理情報学 | 池田 和司, 加藤 博一, 久保 孝富, 日永田 智絵 |
|
title: Eye gaze analysis of experts midwives and novice midwives in assisted childbirth techniques
abstract: In recent years, the number of older births has been increasing due to later marriages, which in turn has led to an increase in high-risk births. Therefore, improvement of midwives' skills is essential for safe childbirth. In a previous research, we analyzed the actual behaviors of midwives and found that they did not follow the textbooks, which may be due to a gap between midwives' awareness and their actual behaviors. The purpose of this study is to objectively clarify the birth attendant skills of skilled midwives and apply them to teaching. We measured the gaze of skilled midwives and novice midwives at a practice site that simulates a simulated birth scene, and analyzed the differences. language of the presentation: Japanese | |||
| 桑原 拓海 | M, 1回目発表 | コンピューティング・アーキテクチャ | 中島 康彦, 林 優一, 張 任遠, KAN Yirong, PHAM HOAI LUAN |
|
title: Spiking k-NN for Human Action Recognition(HAR) using Event based sensor
abstract: Neuromorphic computing is a technology that aims to achieve higher speed and lower power consumption than conventional computing by implementing hardware with Spiking Neural Networks (SNN), which are biologically closer to the human brain than artificial neural networks. It is also possible to approximate k nearest neighbor algorithm using a single layer of SNN. Event-based sensors, which mimic the human eye and convert changes in brightness into spike signals, contribute to reduced information volume and lower power consumption. The combination of event-based sensors and SNN hardware is intended to achieve low power consumption and fast operation for human action recognition. language of the presentation: Japanese 発表題目: イベントベースカメラを用いた、Spiking k-NNでの行動認識 発表概要: ニューロモーフィック・コンピューティングとは、人工ニューラルネットワークよりも生物学的に近い、スパイキングニューラルネットワーク(SNN)をハードウェアで実装することで、通常のコンピューティングより、高速かつ低消費電力を実現しようとする技術です。 また、k近傍法アルゴリズムを、1層のSNNで近似的に実装する事が可能です。 イベントベースセンサーは、ヒトの目を模倣し、輝度変化をスパイク信号に変換するセンサーです。 これにより情報量が少なくなり、低消費電力になります。 イベントベースセンサーとSNNハードウェアを組み合わせることで、低消費電力かつ高速動作する、ヒトの動作認識を目指します。 | |||
| 竜田 明浩 | D, 中間発表 | ネットワークシステム学 | 岡田 実, 林 優一, 東野 武史, DUONG QUANG THANG（客員准教授）, CHEN Na |
|
title: Control and evaluation methods for intelligent reflecting surfaces
abstract: The millimeter-wave and terahertz frequencies are expected to be used in 5G and Beyond 5G mobile communication systems. Comparing with conventional radio waves at lower frequency, they suffer from greater attenuation over distance and susceptibility to blocking, making it difficult to stably cover the required service area. Intelligent Reflecting Surface (IRS) is attracting attention as a method to solve this problem. IRS needs to appropriately control the direction of radio wave reflection from the base station according to changes in the location of the mobile terminal, so it acquires mobile terminal information (location information, etc.) via the base station and determines the direction of radio wave reflection, but the method for controlling this has not yet been established. In this research, we aim to estimate the arrival direction of radio waves into the IRS, propose the autonomous IRS that can reflect radio waves in an appropriate direction, and develop an optimal control method. Therefore, we first constructed an analytical model and computer simulation environment for the mobile communications propagation environment, including IRS, and we verified the validity of the simulation by comparing simulation results using the campus local 5G network with actual measurements. We constructed a simulation environment that is the basis for establishing the IRS optimal control method. language of the presentation: Japanese | |||
| 大岡 冬偉 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基, 林 優一, 妙中 雄三 |
|
title: Automatically managed public-key certificates that guarantee the identity of the domain owner
Abstract: Public Key Infrastructure (PKI) provides trust in the authenticity of public keys to enable secure communication over the Internet. There are two types of public key certificates in use: Domain Validation (DV) certificates and Extended Validation (EV) certificates. DV certificates, which are issued only after confirming that a domain is owned, have the disadvantage of being issued with a low level of trust and can be obtained for fraudulent purposes. On the other hand, EV certificates are issued after human verification of the domain name and the identity of the domain holder, so they have a high level of trust and are only issued to legal entities. However, EV certificates have the disadvantage of being difficult to manage automatically. In recent years, automatic management of certificates has become necessary due to the widely-spread of Let's Encrypt, a service that automatically issues EV certificates, and ACME, a protocol for automatic certificate management. The direction of this research is to realize public key certificates that can be automatically managed to verify the identity of the domain holder in addition to confirming of retention of domain. This will enable the realization of high trust public key certificates that guarantee the identity of the domain holder in an automatically manageable form. Specifically, if the domain holder is an individual, a public personal authentication system such as a My Number card is used to verify the domain holder's identity, and if the domain holder is a corporation, the identity is verified using Verifiable Credentials (VC) technology to issue an Identity Proof VC. This identity proof VC is used to verify the identity of the individual. Finally, the CA issues the proposed public key certificate to the party whose domain retention and identity proof VC can be verified. This research makes it possible to extend the public key infrastructure to issue public key certificates that guarantee identity in a way that allows automatic certificate management. language of the presentation: Japanese 発表題目: ドメイン保持者の身元保証と証明書の自動管理を目的とした公開鍵基盤 発表概要: PKI(公開鍵基盤)はインターネット上において、安全な通信の実現を目的とし、それを実現する為に公開鍵の真正性に対する信頼を提供している。主に用いられている公開鍵証明書にはDV証明書とEV証明書の2種類がある。DV証明書はドメインの保持のみ確認して発行する証明書であり、信頼のレベルが低いため、不正な目的で入手されることがあるというデメリットがあるが、証明書の自動管理が容易というメリットがある。逆に、EV証明書はドメインのに加え、ドメイン保持者の身元等を人間の手によって確認して発行する証明書なので信頼のレベルは高く、法人に対してのみ発行される。しかし、EV証明書は自動管理が難しいというデメリットがある。近年では、Let's EncryptというDV証明書を自動発行するサービスや証明書の自動管理の為のプロトコルであるACMEの普及等により、証明書の自動管理が必要となっている。本研究の方向性として、ドメインの保持に加え、ドメイン保持者の身元の確認を行う自動管理可能な公開鍵証明書の実現を目標とする。これにより、ドメイン保持者の身元まで保証される信頼のレベルの高い公開鍵証明書を自動管理可能な形で実現できる。具体的には、ドメイン保持者が個人の場合はマイナンバーカード等の公的個人認証制度を、法人の利用して身元の確認を行い、Verifiable Credentials(VC)という技術を用いて、身元証明VCを発行する。この身元証明VCにより、身元の確認を行う。最終的に、認証局はドメイン保持の確認と身元証明VCの確認ができる相手に対して今回提案する公開鍵証明書が発行される。この研究により、証明書の自動管理が可能な形で身元保証を行う公開鍵証明書を発行できるように公開鍵基盤を拡張することが可能となる。 | |||
| 角野 匠 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一, 岡田 実, 向川 康博, 藤本 大介 |
|
title: A Study on Determination Method of Noise Level for Masking Measures Against Electromagnetic Information Leakage to Displays
abstract: Display devices are showing information containing confidential data. On the other hand, this information may be leaked due to the screen information not being encrypted depending on the transmission protocol. Eavesdropping threats that reconstruct screen information from unintentional electromagnetic (EM) waves emitted during the display's rendering process have been reported. To counteract the leakage of screen information via EM waves, a method has been proposed to make it difficult to reconstruct screen information during the display's rendering process. As one of the display rendering methods, there is a masking method by adding noise to the screen information; however, the noise level may degrade the user's visibility. This study investigates a method for determining the amount of noise added to screen information as masking against EM leakage of screen information by guaranteeing the user's visibility. language of the presentation: Japanese | |||
| 大貫 和基 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一, 岡田 実, 安本 慶一, 藤本 大介 |
|
title: A Study on Efficient Key Estimation Method by Differential Fault Attack Using Side-Channel Information
abstract: Cryptographic technology is used in many communication devices to protect information security. Cryptography is guaranteed to be theoretically secure, as it takes a lot of time to estimate the secret key. On the other hand, there is a threat of fault injection attacks, which estimate the secret key by physical access to the cryptographic module. A method known as differential fault analysis (DFA) can reduce the search space for secret key candidates. This is achieved by analyzing the differences between correct ciphertext and incorrect ciphertext, which is generated by injecting a fault into a cryptographic module. DFA is known that the secret key can be efficiently estimated by identifying the number of fault bytes. However, identifying the number of fault bytes is difficult. Therefore, this study investigates a method to efficiently estimate the secret key by identifying the number of fault bytes through analysis using side-channel information of the cryptographic module. language of the presentation: Japanese | |||