Researcher List

Laboratories in Department of Information and Computer Science

Nano-Integration Systems Laboratory
Prof. Hiroshi MAKINO

LSI design research in action
LSI design research in action

It is no exaggeration to say that the progress of IT devices such as personal computers, mobile phones, and game consoles is due to the progress of LSIs themselves. Recently, extremely advanced systems such as artificial intelligence have begun to be realized by improving the performance of LSIs. However, while the world is becoming more convenient, LSI design is becoming more and more difficult. In order to solve the increasingly serious problems of LSI design, our laboratory is conducting various researches to achieve higher speed, lower power consumption, and more stable operation of LSIs.

Main Research Topics

  • Operation stabilization of memory circuits in consideration of variation
  • Improvement of LSI operation yield by adaptive voltage optimization
  • Monte Carlo simulation of SRAM operation limits
  • Research on high-efficiency power supply circuits

Intelligent Application Systems Laboratory
Prof. Atsuo OZAKI

Research Topics of Intelligent Application Systems Laboratory
Research Topics of Intelligent Application Systems Laboratory

Most of the large-scale systems, e.g. aerospace system, car traffic system, swarm of pedestrians and so on, are complicated structure. In these systems, to recognize the related situation and to predict the next time situation are required for taking a precise action. We are researching the basic behavior of these complicated systems, and studying an intelligent control technique to construct the decision support system by utilizing following AI (artificial intelligence) technologies.

Main Research Topics

  • Machine leaning technology to precisely recognize the related situation.
  • Multi agent simulation to improve prediction precision.
  • Meta strategy to optimize the system.
  • Parallel and distributed computing to speed up the simulation.

We aim at contributing to realization of the smart society by adapting this decision support system. This society is supposed to be based on smart mobility system, robust crisis management system, and so on, and these are our target applications.

Mathematical and Applied Systems Laboratory
Prof. Yoshinari KAMAKURA

Body temperature power generation system (left) / Simulation to explore how to efficiently extract power from temperature differences (right)
Body temperature power generation system (left) / Simulation to explore how to efficiently extract power from temperature differences (right)

In our laboratory, we are studying about modeling and simulation of semiconductor devices, which are the main components of computers, in order to realize high-performance, high-efficiency, and failure-resistant information systems. For this purpose, we use high performance computing hardware such as GPU and FPGA, as well as machine learning algorithms. In addition, we also develop the systems utilizing the semiconductor devices to learn and understand the requirements from the application side.

Main Research Topics

  • Modeling and simulation of semiconductor devices
  • Modeling and simulation techniques assisted by machine learning
  • Energy harvesting IoT systems powered by thermoelectric generators
  • Development of systems utilizing artificial intelligence

Vehicle Control Laboratory
Associate Prof. Nobuo KOMATSU

Four-rotor helicopter for experiments
Four-rotor helicopter for experiments

In recent years, automatic control technology for moving objects such as automobiles has advanced. In our laboratory, we are studying the position measurement method of moving objects (cars, airplanes, helicoopters, etc.) and automatic guidance technology. We are proposing some methods for position measurement with plural sensors (image sensors, laser sensors, ultrasonic sensors, accelerometers, gyros, etc.). We are also proposing new computer control systems for motors or other actuators so that a moving object travels according to the given path. And, we are conducting experiments with prototype systems to confirm the effectiveness of the proposed methods.

Main Research Topic

  • Position measurement and control of moving objects

System Architecture Laboratory
Associate Prof. Chikako NAKANISHI

Edge AI
Edge AI

In the embedded market, "edge AI" that can be processed independently at the edge (terminal side) is expected due to issues such as operational costs, security, and real-time performance. If AI is installed in each terminal itself, complex processing can be performed on the spot. While there is a strong need for FPGA-based edge AI, it has several problems: it requires expensive FPGA devices, it cannot be implemented well without learning on FPGA-specific networks, and it suffers from accuracy degradation due to quantization.
Image recognition technology has begun to be used in various fields such as face recognition, surveillance and security, image diagnosis in the medical field, defective product determination and fault prediction in factories. We are focusing on these image recognition technologies and are challenging to implement them in edge devices MPSoCs.
Specifically, the FPGA(accelerator) implements the processing that accounts for 80-90% of the performance, while the CPU in MPSoC executes the other processing. the processing executed by the CPU is created using only the standard C library. The accelerator function is part of the inference part, eliminating the need for an expensive FPGA. In addition, the algorithm is designed so that the CPU and FPGA work in concert to make the accelerator effective, so the inference process can be executed quickly.

Main Research Topics

  • Research on Image recognition and object detection using edge AI
  • Co-design of hardware and software
  • Development of accelerator by high-level synthesis and its implementation on FPGA

Neuromorphic Systems Laboratory
Associate Prof. Hirotsugu OKUNO

Robots developed in our laboratory and the future they will open up
Robots developed in our laboratory and the future they will open up

This laboratory has the following two goals. The first goal is to develop efficient systems for visual perception and cognition inspired by the visual nervous systems. The visual nervous systems perceive and recognize visual signals efficiently in terms of both energy and time, while understanding of visual information is one of the most difficult task for conventional computers even though its processing speed is much faster than that of neurons. Learning from the advantages of nervous systems is a promising way to develop noble visual systems. The second goal is to reveal how the efficient visual signal processing is accomplished in the visual nervous systems, including the retina and the visual cortex. We have to reveal it before learning from the visual nervous systems because they remain poorly understood. Understanding the visual nervous systems and developing noble systems for visual perception and cognition inspired by the nervous systems: these two points are the keys of this laboratory.

Main Research Topics

  • Image recognition using artificial intelligence composed of spiking neurons
  • Visual aided control of a small robot inspired by insect vision
  • Image sensor system that is insusceptible to illumination changes
  • Studies on neuromorphic active vision using a binocular robot

Processor Architecture Laboratory
Assistant Prof. Masahito KONISHI

Processor (real) and its design tool
Processor (real) and its design tool

Modern society would not be possible without information devices. Processors are embedded in computers and various other information devices, and are responsible for major processing. In order to make information devices, which can be said to be the foundation of modern society, faster and more convenient, it is extremely important to improve the performance of processors. In order to design a high-performance processor, we consider the architecture of the processors that can execute programs efficiently. We propose and evaluate various speed-up techniques, such as predictive speculative execution and multi-threaded processing, and design high-performance processors.

Main Research Topics

  • Research on branch prediction techniques
  • Research on load latency hiding techniques
  • Multi-threading

Image Processing Laboratory
Assistant Prof. Takao JINNO

Example of our research results
Example of our research results

We research on image processing technology for surveillance or in-vehicle cameras, medical images, crack detection, physiological plant disorder detection, illuminant color estimation and X-ray inspection.
We had engaged to the high dynamic range (HDR) image processing that will be introduced at the end. Recently, we have engaged to detection or visualization weak but useful features in images based on the HDR image processing techniques.
[about the HDR image processing] The dynamic range which can be captured with common cameras and can be displayed with the common monitors is lower than the human eyes. Human eyes can visible the HDR scenes. Capturing the HDR scenes with the common cameras should fuse the multiple exposure images. Displaying the HDR scenes on the common monitors should compress the information of it effectively.

Main Research Topics

  • Visualization of faint features (for X-ray inspection)
  • Color appearance or Color constancy against colored lights
  • Robust Object tracking or Object detection for color changes
  • Multiple illuminant color estimation using a single image

Learning & Advanced Intelligent Systems Laboratory
Assistant Prof. Naoki KOTANI

The humanoid robot used in the research
The humanoid robot used in the research

In our laboratory, we research on machine learning, e.g., reinforcement learning and neural networks, to develop intelligent mechanical systems that learn and evolve like humans. Machine learning is one of the hottest research field of artificial intelligence (AI). One of our research goals is to apply reinforcement learning to make humanoid robots which get adaptive action in environment. Reinforcement learning agents acquire autonomously optimal action in order to maximize the reward which is given by an environment. However, it is very difficult to apply this to robots because reinforcement learning agents based on trial and error need a lot of learning time. In order to solve this problem, we have developed a learning method that reduces the learning time.
We also study food recognition systems and sign language recognition systems using deep neural networks to make our lives healthier and more convenient in our life.

Main Research Topics

  • Research on speeding up reinforcement learning
  • Research on AI-equipped robots
  • Research on recognition systems using deep neural networks