Hello, this is the Public Relations Team of the Matsuo Institute.
The year 2021 will soon be over.
Today, we look back at a summary of Matsuo Lab’s activities in 2021!
1. Papers
This year, the following international conferences and journals have accepted our work.
NeurIPS2021
SPOTLIGHT “Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization
Author：Yusuke Iwasawa, Yutaka Matsuo
Co-Adaptation of Algorithmic and Implementational Innovations in Inference-based Deep Reinforcement Learning
Authors: Hiroki Furuta, Tadashi Kozuno, Tatsuya Matsushima, Yutaka Matsuo, and Shixiang Shane Gu
◆ICML2021
Policy Information Capacity: Information-Theoretic Measure for Task Complexity in Deep Reinforcement Learning
Authors: Hiroki Furuta, Tatsuya Matsushima, Tadashi Kozuno, Yutaka Matsuo, Sergey Levine, Ofir Nachum, and Shixiang Shane Gu
◆ICLR2021
Group Equivariant Conditional Neural Processes
Authors： Makoto Kawano, Wataru Kumagai, Akiyoshi Sannai, Yusuke Iwasawa, and Yutaka Matsuo
Deployment-Efficient Reinforcement Learning via Model-Based Offline Optimization
Authors: Tatsuya Matsushima, Hiroki Furuta, Yutaka Matsuo, Ofir Nachum, and Shixiang Shane Gu
◆EMNLP2021
Findings “Subformer: Exploring Weight Sharing for Parameter Efficiency in Generative Transformers
Authors: Machel Reid, Edison Marrese-Taylor, Yutaka Matsuo
AfroMT: Pretraining Strategies and Reproducible Benchmarks for Translation for 8 African Languages
Author：Machel Reid, Junjie Hu, Graham Neubig, Yutaka Matsuo
◆ACL-IJCNLP 2021
Findings “LEWIS: Levenshtein Editing for Unsupervised Text Style Transfer. “
Authors: Machel Reid and Victor Zhong
◆UAI2021
Improved Generalization Bounds of Group Invariant / Equivariant Deep Networks via Quotient Feature Spaces
Authors: Akiyoshi Sannai, Masaaki Imaizumi, Makoto Kawano
Machine Learning (Springer)
Information-theoretic regularization for learning global features by sequential VAE
Author： Kei Akuzawa, Yusuke Iwasawa, Yutaka Matsuo
Web Intelligence
Graph-based Knowledge Tracing: Modeling Student Proficiency Using Graph Neural Network.
Author： Hiromi Nakagawa, Yusuke Iwasawa, Yutaka Matsuo
◆L4DC
Estimating Disentangled Beliefs about Hidden State and Hidden Task for Meta-Reinforcement Learning.
Author： Kei Akuzawa, Yusuke Iwasawa, Yutaka Matsuo
Advanced robotics
A survey of multimodal deep generative models, Advanced Robotics
Author： Masahiro Suzuki, Yutaka Matsuo
◆Artificial Intelligence Society of Japan
Recommendation System Using Conditional Variational Autoencoder Considering Action Time
Author： Jun Hozumi, Yusuke Iwasawa, Yutaka Matsuo
◆Information Processing Society of Japan
Out-of-Distribution Detection Using Simultaneous Class and Geometric Transform Probabilities with Deep Neural Networks, Transactions of Information Processing Society of Japan
Author：Hirono Okamoto, Masahiro Suzuki, Yutaka Matsuo
◆Information Processing Society of Japan Transaction Digital Practices
Indexation of Financial and Economic Reports Using the LSTM Model”
Author： Hiroki Yamamoto, Keiichi Ochiai, Masahiro Suzuki, Yutaka Matsuo
2. Lecture
Matsuo Lab offers various lectures on deep learning, data science, and web engineering.
In FY2021, two new endowed courses, “AI Management Endowed Course” and “World Model Simulator Endowed Course,” were established, bringing the total number of students to 3,200 for the year! (160% compared to last year)
The AI Management Endowed Chair was established and 14 lectures were held for graduate students of the University of Tokyo starting in June.
After learning about AI x management from practitioners on a vertical (industry-specific) and horizontal (cross-industry) basis, participating students were divided into groups to discuss and present AI solutions to solve social issues.
The program for working adults will begin in January 2022, and the next lecture for Todai graduate students is scheduled for April 2022.
◆World Model Simulator Endowed Chair opened “World Model and Intelligence” in January 2010.
An endowed lecture course on “World Models,” a research focus area of the Matsuo Research Institute, has been established.
The “World Model and Intelligence” lecture started in January 2010 with over 160 applicants.
◆Summer School offers lectures on “Image Recognition” as a new theme
In addition to natural language processing, reinforcement learning, and generative modeling, a fourth theme, “Image Recognition,” was offered, attracting more than 400 participants.
Deep Reinforcement Learning,” which was previously offered at the Summer School, was held at the Autumn Seminar.
☆ March 2022: Spring Seminar begins.
Information will be released around January 2022.
We will announce on Twitter ( @Matsuo_Lab ) when we start accepting applications.
◆Graduate School of Engineering “Entrepreneurship Education and Design Endowed Chair” established.
It was established to design entrepreneurship education for the entire graduate school with the aim of further creating university startups. (Photo taken at the opening press conference)
Prof. Matsuo, together with Prof. Sakata, Prof. Kagami, and Associate Prof. Tanaka, is planning and supervising the program.
Since October, 13 lectures have been held and more than 120 students have studied under the theme of Deep Tech x Global Entrepreneurship.
A program for Komaba students in their first and second year of undergraduate studies is scheduled to begin in April 2010.
3. Various study groups and circle activities
Matsuo Lab conducts various study groups and circle activities for researchers and students to implement the latest papers.
DLHacks, a training ground for putting the latest papers on Deep Learning into programming code, has resumed. It is held regularly every week (Friday) from 6:00 pm.
Anyone who has completed the lectures hosted by the Matsuo Lab (and who is a current student excluding working adults) is welcome to participate.
If you would like to improve your technical skills, please apply through the regular recruitment on Slack for alumni!
◆ Robocon TeamPlaces Second at WRS2020
We won 2nd place in the WRS (World Robot Summit 2020) Partner Robot Challenge (Real Space)!
The Robocon team has formed a subgroup of the Matsuo Lab called TRAIL (Tokyo Robot And Intelligence Lab).
If you are interested, please visit the TRAIL website!
WRS on the day of the event (YouTube streaming): https://m.youtube.com/watch?v=ElUb8bfSC34
4. entrepreneurship support
◆”Matsuo Lab Entrepreneurship Quest” beta site opens.
The “Matsuo Lab Entrepreneurship Quest,” an AI startup entrepreneurship training program supervised by Professor Matsuo, has been launched, and two new startups have been born from the first group of graduates from the Matsuo Lab.
The beta version of the website was opened at the end of December. (The photo shows a visit to ACES in the 3rd stage.)
◆Three startup companies are born from Matsuo Lab.
Matsuo Lab startups are companies founded by or with the support of Matsuo Lab graduates that have been selected for their growth potential in terms of both technology and business capabilities, and for their commitment to the Matsuo Lab philosophy and to the training of their successors.
In 2021, three new startups were born for the first time in two years, bringing the total number of startups to 12. Please click here to learn more about Matsuo Lab startups.
5. Recruitment of new members
Matsuo Lab is always looking for new members to join our expanding team! Please see Wantedly for current openings.
https://platform.wantedly.com/projects/819172https://platform.wantedly.com/projects/819169
https://platform.wantedly.com/projects/819170https://platform.wantedly.com/projects/819171
These are the main news items of the Matsuo Institute in 2021!
Please have a happy New Year.
We look forward to working with you in the New Year.
