• Home
  • News
  • Matsuo Lab., The University of Tokyo] Report on Activities in 2022
  • Matsuo Lab., The University of Tokyo] Report on Activities in 2022

    Hello, this is the Public Relations Team of the Matsuo Institute.

    The gingko trees on Hongo Campus have lost their leaves and turned into a winter scene.
    With only a few days left in 2022, we take a look back at the activities of the Matsuo Institute this year.

    1. Publications and Activities

    This year, the following international conferences and journals, including ICLR and NeurlPS, have accepted our work.

    ◆ICLR2022 [Spotlight]
    Generalized Decision Transformer for Offline Hindsight Information Matching
    Author: Hiroki Furuta, Yutaka Matsuo, Shixiang Shane Gu

    NeurIPS 2022
    Langevin Autoencoders for Learning Deep Latent Variable Models
    Author: Shohei Taniguchi, Yusuke Iwasawa, Wataru Kumagai, Yutaka Matsuo

    Large Language Models are Zero-Shot Reasoners
    Author: Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, Yusuke Iwasawa

    EMNLP 2022
    Learning to Model Editing Processes
    Author:Machel Reid, Graham Neubig

    M2D2: A Massively Multi-domain Language Modeling Dataset
    Authors: Machel Reid, Victor Zhong, Suchin Gururangan, Luke Zettlemoyer

    ◆NAACL 2022 (main)
    PARADISE: Exploiting Parallel Data for Multilingual Sequence-to-Sequence Pretraining
    Authors: Machel Reid and Mikel Artetxe

    A Few Thousand Translations Can Go a Long Way: Leveraging Pre-trained Models for African News Translation
    Authors: David Ifeoluwa Adelani, Jesujoba Oluwadara Alabi, Machel Reid

    IJCAI-ECAI 2022 (short)
    Robustifying Vision Transformer without Retraining from Scratch by Test-Time Class-Conditional Feature Alignment
    Authors: Takeshi Kojima, Yusuke Iwasawa, and Yutaka Matsuo

    ◆ICDL 2022
    Brain-inspired Probabilistic Generative Model for Double Articulation Analysis of Spoken Language
    Authors: Akira Taniguchi, Maoko Muro, Hiroshi Yamakawa, and Tadahiro Taniguchi

    ◆MMMM 2023
    Improving the Robustness to Variations of Objects and Instructions
    Author: Kazutoshi Shinoda,Yuki Takezawa, Masahiro Suzuki, Yusuke Iwasawa, Yutaka Matsuo

    ◆ACL 2022 (workshop)
    On the Impact of Data Augmentation on Downstream Performance in Natural Language Processing
    Authors: Itsuki Okimura, Machel Reid, Makoto Kawano, Yutaka Matsuo

    Neural Networks
    A Whole Brain Probabilistic Generative Model: Toward Realizing Cognitive Architectures for Developmental Robots
    Author: Tadahiro Taniguchi, Hiroshi Yamakawa, Takayuki Nagai, Kenji Doya, Masamichi Sakagami, Masahiro Suzuki, Tomoaki Nakamura, Akira Taniguchi

    Remote Sensing
    Detecting Object-level Scene Changes with Viewpoint Difference using Graph Matching
    Author: Kento Doi, Ryuhei Hamaguchi, Yusuke Iwasawa, Yutaka Matsuo, Masaki Onishi, Ken Sakurada

    Frontiers in Robotics and AI
    Conveying Intention by Motions With Awareness of Information Asymmetry
    Author: Yosuke Fukuchi, Masahiko Osawa, Hiroshi Yamakawa, Tatsuji Takahashi, Michita Imai

    IEEE Access
    Explaining Intelligent Agent’s Future Motion on the Basis of Vocabulary Learning With Human Goal Inference.
    Authors:Fukuchi, Y., Osawa, M., Yamakawa, H., & Imai, M.

    Pattern Recognition Letters
    Face-mask-aware Facial Expression Recognition based on Face Parsing and Vision Transformer
    Author: Bo Yang, Yusuke Iwasawa, Yutaka Matsuo

    Deep Learning Pipeline for Spotting Macro- and Micro-expressions in Long Video Sequences Based on Action Units and Optical Flow
    Author: Bo Yang, Jianming Wu, Kazushi Ikeda, Gen Hattori, Masaru Sugano, Yusuke Iwasawa, Yutaka Matsuo

    Advanced Robotics
    World robot challenge 2020 – partner robot: a data-driven approach for room tidying with mobile manipulator
    著者:Tatsuya Matsushima, Yuki Noguchi, Jumpei Arima, Toshiki Aoki, Yuki Okita, Yuya Ikeda, Koki Ishimoto, Shohei Taniguchi, Yuki Yamashita, Shoichi Seto, Shixiang Shane Gu, Yusuke Iwasawa, Yutaka Matsuo.

    New Generation Computing
    Robustifying Vision Transformer Without Retraining From Scratch Using Attention Based Test-Time Adaptation
    Author: Takeshi Kojima, Yusuke Iwasawa, Yutaka Matsuo

    ◆Transactions of the Japanese Society for Artificial Intelligence
    A Proposal of Scene Interpretation Method Using Transformer and Self-Supervised Learning
    Author:Yuya Kobayashi, Yudai Suzuki, Yutaka Matsuo

    M3IL: Multi-Modal Meta-Imitation Learning
    Author: Xin Zhang, Tatsuya Matsushijma, Yutaka Matsuo, Yusuke Iwasawa

    ◆Interpretation Processing Society of Japan
    Fixing the train-test objective discrepancy: Iterative Image Inpainting for Unsupervised Anomaly Detection
    Author: Hitoshi Nakanishi, Masahiro Suzuki, Yutaka Matsuo

    Forecasting Production Indices Using a Deep Learning Model with Inter-industry Trade Structure
    Author: Hiroki Yamamoto, Keiichi Ochiai, Masahiro Suzuki, Yutaka Matsuo

    ◆The Institute of Electronics, Information and Communication Engineers
    A Proposal of Detection Method for DeepFake Video Using Face Angle Information
    Author: Satoshi Kageyama, Masahiro Suzuki, Keiichi Ochiai, Yutaka Matsuo

    <Other Reports

    ◆Organized Session (OS) “ World Models and Intelligence” at the 2022 Japanese Society for Artificial Intelligence (JSAI) National Convention
    At this year’s JSAI National Convention, the Matsuo Lab conducted an OS on world models promoted by the JSAI, “World Models and Intelligence.
    This year was the third time this OS has been held, with the following objectives

    The aim of the workshop is to discuss the methodology of modeling the external world using deep learning and the importance of world models in various AI problems.
    Through the concept of world models, we aim to provide a forum for interdisciplinary discussions in collaboration with researchers not only in artificial intelligence but also in robotics, cognitive science, neuroscience, and other fields.

    For more information

    ◆External lecture meeting organized by Matsuo Lab.
    Lectures by invited lecturers were held for the purpose of research discussion.

    (Lecture 1)
    Speaker: Large Language Models: What will happen next?
    Speaker: Keisuke Sakaguchi (Associate Professor, Tohoku University)

    (Lecture 2)
    Title: Augmenting reinforcement learning with language
    Speaker:Andrew Lampinen (Senior Research Scientist at DeepMind)

    ◆Robot team is preparing for Robocon.
    The domestic RoboCup competition will be held jointly with JSK in March.
    Preparations are underway for the RoboCup World Championship in Bordeaux, France in the summer of 2023.

    Click here to watch a video introducing the ▼Robot Team

    https://youtube.com/watch?v=FFhiQ7z3zfc%3Ffeature%3Doembed

    ◆Added ” Brain-inspired Intelligence ” as a research field.

    At Matsuo Lab, we are focusing on general-purpose artificial intelligence research with the vision of “creating intelligence”.
    We believe that it is essential to expand our research into the area of NeuroAI in order to study intelligence from an engineering perspective,
    Dr. Hiroshi Yamakawa is leading our research.

    2. Professor Matsuo’s Activities

    ◆Prof. Matsuo attended this year’s conference as a member of the “Council for the Realization of New Capitalism”.

    The Matsuo Institute has been contributing to society from a variety of angles based on its activities, including the ” Grand Design” and the Draft Five-Year Startup Development Plan,
    The Matsuo Institute is making multifaceted contributions to society based on its activities.

    3. Lecture

    ◆Lecture attendance exceeded 4,700 per year.

    The Matsuo Institute offers a variety of courses, including the ” Global Consumer Intelligence Endowed Chair,” the ” Global Model Simulator Endowed Chair, ” and the “AI Management Endowed Chair,
    The Matsuo Institute offers a variety of courses, including the “Global Consumer Intelligence Endowed Chair,” the “World Model Simulator Endowed Chair,” and the ” AI Management Endowed Chair.

    New courses such as ” Lottery Hypothesis and Its Development ” and ” Financial Market Transactions and Machine Learning ” were also launched this year.

    ◆Implementation of the AI Management Endowed Summer Boot Camp for students.
    The program for students was changed to an intensive summer boot camp format this year.
    The number of applications far exceeded the maximum number, and the program was held with 48 students selected for the program.

    ◆Registration for AI management donation course for working professionals is now open.

    The lecture will cover the fundamentals of AI, a core digital technology, and its application in various industries,
    The lecture will also provide practical information to deepen your understanding of the use of technology in a variety of industries.
    To register, please click here. Application deadline: Thursday, January 12, 2023

    ◆The Metaverse Engineering Department was opened at the University of Tokyo’s Faculty of Engineering.

    Matsuo Lab is mainly in charge of reskilling courses. Corporate membership is always welcome.

    4. Various study sessions

    ◆DLHacks, weekly DL roundtables

    DLReadings, a weekly rotating group of people who read and present the latest papers on deep learning, and
    DLHacks,” a weekly training session for translating the latest papers on deep learning into programming code.

    Any alumni (and current students excluding working adults) of the lectures hosted by Matsuo Lab are welcome to participate.
    If you would like to improve your technical skills, please apply through the regular recruitment on Slack for alumni.

    4. entrepreneurship support

    ◆Matsuo Lab. launches one start-up company.

    What is a Matsuo Lab Startup?
    Startups are companies founded by or with the support of Matsuo Lab graduates that have been recognized as having growth potential in terms of both technology and business capabilities,
    The startups are selected companies that have been founded by or with support from the Matsuo Institute, and that have been recognized as having growth potential in terms of both technology and business capabilities, and that share the philosophy of the Matsuo Institute and work together to nurture the next generation.

    This year, ” EQUES ” was born as a start-up from Matsuo Lab, bringing the total number of companies to 13.
    The following companies are the only ones officially recognized as “start-ups from Matsuo Lab.

    Start-up from Matsuo Lab.

    ◆Matsuo Lab. and one start-up company from a technical college is born.

    What is “Matsuo Lab Startups from Technical Colleges”?
    The “Matsuo Lab/Koto College Startups” are a group of startup companies founded by or with the support of a technical college graduate that have been selected for their growth potential in terms of both technology and business capabilities, and that share the philosophy of the Matsuo Lab,
    The “Matsuo Lab/SCE Startups” are a group of startup companies founded by or with support from the Matsuo Lab, which have been selected for their growth potential in terms of both technology and business capabilities, and which share the philosophy of the Matsuo Lab and work together to foster the next generation.

    This year, ” integrAI ” was born as a startup from Matsuo Lab and technical college, bringing the total number of companies to two.
    The following companies are the only ones officially recognized as “startups from Matsuo Lab and Colleges of Technology.

    Matsuo Lab., a start-up from a technical college

    ◆”Matsuo Lab Entrepreneurship Quest” begins its third term.

    The third term of “Matsuo Lab Entrepreneurship Quest,” an AI startup entrepreneurship training program supervised by Professor Matsuo, has begun.
    Applications are open to all university grades, and 59 students have participated in this term.
    Applications for the fourth term are also scheduled to begin in January 2023.

    4. recruitment of new members

    Matsuo Lab is looking for researchers, engineers, and many other positions.
    Recruitment information sessions are held twice a month.
    If you are interested in learning more about Matsuo Lab, please join us!

    ▷Click here to apply for a briefing session
    ▷Click here for available positions
    (Researcher / Technical Consultant / Lecture Program Manager, etc.)

    In order to provide you with more transparent information about our laboratory environment, we are also working hard to disseminate information.
    Please read the following articles.

    ◆Interview with Yutaka Matsuo, President and CEO (Part 1): Toward the Realization of an Ecosystem on Par with Silicon Valley

    ◆Matsuo Lab’s Thoughts on Basic Research: Interview with Yutaka Matsuo, President (Part 2)

    ◆What is the “World Model”? Why Matsuo Lab. is promoting research to realize intelligence. (Part 1)

    Serious challenge to realize ◆Intelligence. What is the research environment of the Matsuo Laboratory, which has a multifaceted perspective? (Part 2)

    These are the main news items of the Matsuo Institute in 2022!

    Please have a happy New Year.
    We look forward to working with you in the New Year.