AIP Network Lab

  • Font-size
  • Large
  • Middle
  • Small

AIP Acceleration Research

"AIP Acceleration Research" aims to maximize the research result as an AIP Network Laboratory by supporting newly-proposed research projects based on the excellent research results among the laboratory.

AIP Acceleration Research FY 2020

Operation Knowledge Acquisition and Transfer of Biology Experiments

Research Director

Yoichi Sato Professor, Institute of Industrial Science, The University of Tokyo

Collaborators

Toutai Mitsuyama Team Leader, Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology
Kitani Kris M. Professor, The Robotics Institute School of Computer Science Associate Research, Carnegie Mellon University

Description

We will work on activity recognition and skill modeling from video recordings of biological experiments carried out by lab technicians. In particular, we will develop methods for recognizing and tracking actions and tools involved in the experiments by using videos captured by various types of wearable cameras such as head-mounted or wrist-mounted cameras. We will also develop methods for modeling activity skills of experienced lab technicians and transferring them to biological experiment robots.

Lightweight Cryptoprocessor Architecture for Tiny IoT Edge Devices

Research Director

Yuko Hara-Azumi Professor, School of Engineering Associate, Tokyo Institute of Technology

Collaborators

Yang Li Professor, Graduate School of Informatics and Engineering Associate, The University of Electro-Communications

Description

Side-channel attacks that attempt to obtain personal information via physical information (e.g., power consumption) leaked from IoT edge devices are becoming serious by the advancing AI technologies. This project addresses small, low-power, and secure IoT system developments that can increase the physical security of IoT edge devices. For the realization of the secure IoT society, we develop an integrated system, which combines a novel cryptoprocessor architecture and lightweight cryptographic software implementation, on an IoT-oriented security evaluation platform assuming AI-based attacks.

Construction of High-accurate Prediction Model from Limited Supervised Data

Research Director

Tatsuya Harada Professor, Research Center for Advanced Science and Technology, The University of Tokyo

Collaborators

Masashi Sugiyama Team Leader, RIKEN Center for Advanced Intelligence Project

Description

The purpose of this project is to realize theories and algorithms for learning a high-accurate prediction model from only a small number of labeled or weakly labeled data and to develop an automatic calculation system for them. With the success of machine learning in recent years, prediction accuracy has been dramatically improved. Still, the problem remains that enormous amounts of labeled data are needed to obtain high prediction accuracy, which requires a great deal of human labor. Therefore, in this research, we will tackle this problem from the viewpoints of weakly supervised learning, knowledge transfer, and automatic learning for them.

Cosmology with Big Astronomcal Data Using Innovative Image Analysis Methods

Research Director

Naoki Yoshida Professor, University of Tokyo Institutes for Advanced Study,

Collaborators

Naonori Ueda Head of Ueda Research Laboratory/NTT Fellow, NTT Communication Science Laboratories,
Masaomi Tanaka Associate Professor, Graduate School of Science,
Shiro Ikeda Professor, The Institute of Statistical Mathematics,
Kiyotomo Ichiki Associate Professor, Graduate School of Science,
Takahiro Nishimichi Associate Professor, Yukawa Institute for Theoretical Physics,

Description

We analyse a large set of data from Subaru HSC and Tomo-e Gozen. We detect and classify automatically distant supernovae and rare transient phenomena. To this end, we develop novel methods for analysing peta-bytes imaging data and for data compression. We aim at detecting gravitational lensing signals from images of millions of galaxies. Finally, we develop a fast emiulator that calculates statistical quantities that characterize the large-scale distribution of matter in the Universe.