Machine Learning Tokyo
Founders: Suzana Ilic, Yoovraj Shinde
Founded in: July 2017, Tokyo, Japan
Registered nonprofit 一般社団法人 since: May 2019
Representative Director: Suzana Ilić
Board of Directors: Suzana Ilić, Yoovraj Shinde, Alisher Abdulkhaev, Dimitris Katsios
Members: 7,100 (October 2020)
Machine Learning Tokyo:
Artificial Intelligence in Japan
Machine Learning Tokyo (MLT) is an award-winning nonprofit organization 一般社団法人 based in Japan. MLT is dedicated to democratizing Machine Learning through open education, open source and open science. We support an international research- and engineering community of 7,000 members.
Our mission is to grow and sustain an inclusive, collaborative and engineering focused environment for students and professionals. We provide opportunities to study and work on Machine Learning and Deep Learning projects. Since its founding MLT held more than 100 technical Machine Learning, Deep Learning or Math workshops, talks and study sessions. We provided hundreds of hours of study and implementation for free.
Open Education | Open Source | Open Science
Open Education – MLT held more than 200 AI workshops, study sessions and talks with thousands of participants in Tokyo and remote with participants from all over the world. Our events are inclusive and with an open education mindset, individuals can attend all events free of charge.
Open Source – Several volunteer teams within the MLT community are working on Machine Learning, Deep Learning, Reinforcement Learning and Robotics projects, including substantial work that has been done in the field of AI for Social Good. All projects are hosted on the public Machine Learning Tokyo GitHub Organization; code bases and repositories are published as open source projects.
Open Science – MLT teams have published research papers at international AI conference workshops and we’re continuously collaborating with Universities and Research Institutes in Japan to support open science and researchers with diverse academic backgrounds, including the University of Tokyo, Tokyo Institute of Technology and RIKEN CBS.
The MLT community is an active community of students, research and industry professionals, such as Machine Learning Engineers, Research Scientists, Data Scientists, Software Engineers, Roboticists and Hardware Engineers, etc. It consists of 7,000 members on Meetup and 30,000 followers on Social Media. The demographics show that 60-70% are highly skilled technical professionals, ranging from entry level to senior and director level.
Source: LinkedIn Analytics for Machine Learning Tokyo (Jan 2020)
MLT holds 3-5 Machine Learning physical events per month, during COVID-19 we increased the number of meetups and are currently holding 15 online study sessions per month. These range from hands-on workshops (algorithms, models, production-level Deep Learning, etc.) and regular study sessions (ML Math, Recommendations Systems, Edge AI, etc.) to talks and panels that are aimed to be knowledge-sharing events for the general public, and complementary to our highly technical events for a research- and engineering audience.
Deep Learning workshops
Our MLT engineers (all volunteers) create resources and materials for specific topics to teach other engineers and researchers about cutting-edge AI technology and state-of-the-art methods. One of our regular workshops goes through Convolutional Neural Networks that are used for Computer Vision Tasks like Object Detection, Recognition and Segmentation. This technology has had a huge impact on different verticals and application areas, such as Health Care (cancer, pneumonia, etc. detection, …), Self-Driving Cars (person, lane, etc. detection, …) and many other fields.
Our events were held in front of thousands of participants – engineers and researchers, some of which were recorded and published online as a video lecture series to reach a larger and broader audience.
One of our most recent study series is dedicated to Machine Learning Math. More than 1,000 people from all over the world signed up for our remote MLT Machine Learning Math Reading Sessions もくもく会 – bi-weekly reading sessions and discussions around mathematical concepts that are relevant for Machine Learning and Deep Learning such as Linear Algebra, Matrix Decompositions, Vector Calculus and more. Apart from sessions that are based in and hosted from Japan, our Math Reading Sessions led to amazing collaborations and new leadership teams that are hosting sessions in the US, Germany, the UK and India.
Other regularly hosted MLT study sessions include Recommendation Systems, Edge AI and Embedded Machine Learning Systems, Natural Language Processing: Paper Reading Sessions, Deep Learning: Generative Models, etc.
Talks & Panels
In the past two years we were fortunate to welcome some of the world’s leading Deep Learning experts in Tokyo to give talks for Machine Learning Tokyo. We organized a talk and Q&A session with Francçois Chollet – Research Scientist at Google AI and the Creator of Keras, a Deep Learning API with more than 300,000 active users. The talk was held in one of the biggest lecture halls at the University of Tokyo and we were able to accommodate 500 people in the audience.
Other guest speakers include Anima Anandkumar (Prof. at Caltech, Machine Learning Research Director at NVIDIA), David Ha (Research Scientist and Head of Google Brain Tokyo), AI Pioneer Andrew Ng (remote – Coursera, deeplearning.ai), Josh Gordon (Google AI New York), etc.
Women in Machine Learning
The event series MLT Women in Machine Learning is one of our Diversity & Inclusion initiatives. Twice per year we hold this event with multiple female speakers in Machine Learning and Data Science and an audience of up to 150 people. We aim to create a community culture that is collaborative, inclusive and empowering and free of intimidation, bias and discrimination. These events are specifically designed to support members from underrepresented groups, to provide visibility, empowerment and mentorship. For previous events we have partnered with Google Japan, Mercari, and other companies that are actively supporting Diversity and Inclusion in Tech and AI.
Images: MLT Women in Machine Learning event with more than 100 attendees, January 2020 at Google Japan.
MLT is a hands-on community, consisting of about 70% technical Machine Learning talent. We provide the opportunity to experiment and explore new technology, including software and hardware, at MLT Hackathons and challenge the community to work creatively and collaboratively to solve technical challenges in novel ways and deploy and present them as use cases.
Our most recent event “Deep Perception Hackathon” brought together engineers, developers, researchers and designers to put Deep Learning models into VR and AR. The first day was dedicated to workshops and talks with experts in the field such as Prof. Mark Billinghurst (Empathic Computing Lab, University of South Australia) who gave an intro to Augmented Reality or Naoji Taniguchi (CEO Holoeyes Inc.) who showed an interactive demo of their VR technology and how it is revolutionizing healthcare communication.
The teams worked different cutting-edge VR hardware like HTC Vive, Oculus RIFT, Windows Mixed Reality Headset, Microsoft HoloLens and complementary hardware like the Jetson Nano. The presented demos ranged from creative VR Applications involving BigGANs, AR Applications using Object Detection and many more.
The 2-day hackathon was held at Mistletoe of Tokyo and organized and mentored by Machine Learning Tokyo (Deep Learning), the Digital Nature Group at the University of Tsukuba (VR, AR) and ON-1 Tokyo (Creative Design Community).
MLT continues to make educational efforts, though simultaneously, we are engaged in multiple AI projects. The following is a selection of MLT projects:
Stanford Deep Learning Translation: EN-JP
An MLT team of 12 bilingual volunteers translated Stanford’s CS 230 Deep Learning course notes from English into Japanese to make high-quality education from top-level Universities accessible and support Japanese speaking ML beginners and students.
- 畳み込みニューラルネットワーク: http://stanford.io/2LRE8py
- リカレントニューラルネットワーク: http://stanford.io/30R34C1
- アドバイスやコツ: http://stanford.io/2ok3O5h
EN-JP Machine Learning and Deep learning Lexicon is a spin-off project that provides a sorted list of key terms and phrases in both languages.
Contributors: Yoshiyuki Nakai, Yuta Kanzawa, Hideaki Hamano, Tran Tuan Anh, Nao Takatoshi, Kamu, Rob Altena, Oniki, Suzana Ilić
Generative Adversarial Networks for creative text generation
For this work we explored creative generative models for text (poems, lyrics, metaphors) with three main components:
(1) Pre-trainig on large-scale creative text (Gutenberg novels),
(2) implement AWD-LSTM and Transformer XL-based Generative Adversarial Networks with a Discriminator feedback that evaluates if a piece of text is creative or non-creative and
(3) substitute maximum likelihood optimization (MLE) and optimize for creative rather than the most likely tokens to elicit unique word combinations.
Just as the results of a human creative process, not all outputs of our generative models are interesting. But we do see AI Fragments, that are unusual, strange, funny, deep, and definitely creative. Some curated creative outputs are displayed on our website.
This research project was accepted as a workshop paper at NeurIPS 2019: Machine Learning for Creativity and Design and was presented in December 2019 in Vancouver, Canada.
Saeed, A., Ilic, S., Zangerle, E. (2019): Creative GANs for generating poems, lyrics, and metaphors. (accepted to the NeurIPS 2019 Workshop: Machine Learning for Creativity and Design)
Deep Learning-based Chatbot in the browser
How do we design and build machines with human-like traits? In this research project we explored the idea of personality-based dialogue systems from a conversation design perspective.
For many years artificial intelligence researchers have been investigating how to design and build machines that are not only able to understand and reason, but to perceive and express emotions. A more recent stream of NLP and machine learning research is dedicated to generative systems that model human characteristics as a key component for natural human-machine conversations and interactions. Rather than being task-oriented virtual assistants, those systems have personalities or identities and display opinions and emotions in open-domain settings. In this work, we focus on training an end-to-end system on a carefully crafted dataset that reflects human-like traits, emotion, humor and sarcasm.
This research project was accepted as artwork to the online gallery of the NeurIPS 2018 Workshop: Machine Learning for Creativity and Design.
Open Source: https://github.com/Machine-Learning-Tokyo/seq2seq_bot
Contributors: Suzana Ilić, Reiichiro Nakano
Kuzushiji Optical Character Recognition
[Junction Tokyo 2019 – Hackathon] Before Meiji, when Japanese characters weren’t standardized, people wrote books using a cursive script called Kuzushiji. Today, most Japanese natives cannot read Kuzushiji. This means there are over a thousand years’ worth of books (~3 million unregistered books and a billion historical documents) that are inaccessible to the general public. Our solution is a web application serving a Kuzushiji Optical Character Recognition (OCR) system. The web application lets users upload images or take pictures, and detects the location of each character and classifies them.
Open Source: https://github.com/Machine-Learning-Tokyo/kuzushiji-lite
Contributors: Asir Saeed, Reiichiro Nakano
MLT x 2020: Face, age, gender, emotion detection
[Junction Tokyo 2019 – Hackathon] More than half a million people are expected to come to Tokyo for the Olympics 2020. In less than 48 hours our team built a scalable system for face detection and count, age, gender and emotion prediction for managing crowds, making personalized recommendations for users optimizing marketing campaigns and ad placement (e.g. ads on screens based on average age, gender). We combined 3 Deep Learning Models and 3 APIs (Twitter, Google Maps, Google Translate). This helps to facilitate Safety, Efficiency and Business Value for the Tokyo Olympics 2020.
Open Source: https://github.com/Machine-Learning-Tokyo/MLTx2020
Contributors: Sai Htaung Kham, Alisher Abdulkhaev, Ben Ioller, Dimitris Katsios, Suzana Ilić
Tactile Patterns for visually impaired and blind people
This is a collaborative project between MLT and Prof. Tune Kamae, a former physics professor (University of Tokyo, Stanford University) who has been engaged for years in supporting, teaching and creating learning resources for visually impaired and blind people. MLT is supporting his efforts. The goal of this project is to convert photo to Tactile Image that can be printed with Tactile Printer on Swell Paper to assist visually impaired person in seeing the photo.
- Input image and output patterns (left)
- Blind people are testing the tactile patterns representing the original image (right).
Open Source: https://github.com/Machine-Learning-Tokyo/tactile_patterns
Contributors: John Lau, Suzana Ilić
An MLT Team of 15 engineers and researchers are dedicated to projects related to embedded Machine Learning and Robotics. One of our projects is MLT x Agritech, where we investigate new ways of improving and automating agricultural processes with cutting-edge technology, from hardware to software.
In July, the team visited the Hackerfarm, a community of farmers and engineers in a remote area of Chiba, 2-3 hours from Tokyo. The team talked to local people, technologists and farmers, that explained their efforts and pain points. People in rural areas of Japan are getting too old or sick to take care of the vast lands and fields. They are not able to maintain those anymore and with that, precious knowledge that was built up over generations gets lost. There is a need to develop technologies and automation to support those farmers.
First steps included looking at an existing work in progress, a suspended robotic system for automating tasks, and we discussed further opportunities for IoT and Machine Learning projects to be developed. In the initial research phase, we started experimenting with Deep Learning models that can run on small power-efficient micro-controllers. Soon we realized that this could be an impactful technology, not only for Japan, but especially for developing countries that suffer from severe constraints regarding infrastructure, financial and other resources
After the experimental setup was successful we documented our results. We presented a quantized deep neural network deployed on a low-power edge device, inferring learned motor-movements of a suspended robot in a defined space. This served as the fundamental building block for the original setup, a robotic system for farms or greenhouses aimed at a wide range of agricultural tasks.
Deep learning on edge devices and its implications could have a substantial impact on farming systems in the developing world, leading not only to sustainable food production and income,
but also increased data privacy and autonomy, with the following advantages compared to classical systems:
- Low latency: Fast on-device inference
- Privacy: Data is processed on-device
- Connectivity: Fully offline
- Power consumption: Highly power efficient
- Low-cost: Purchase and working mode
This research project was accepted as a workshop paper at NeurIPS 2019: Machine Learning for the Developing World (ML4D) and was presented in December 2019 in Vancouver, Canada.
Sinha, A., Kumar, N., Mohanan, M., Rahman, M., Quemener, Y., Mim, A., Ilić, S. (2019): Quantized deep learning models on low-power edge devices for robotic systems
MLT’s recognition at a global scale includes winning the Rakuten Technology & Innovation Silver Award in 2019, and presenting our research efforts at international conferences.
- NeurIPS (Montreal, December 2018)
- Data Science Hong Kong (Hong Kong, Jan 2019)
- Poster presentation at Pytorch Developer Conference (San Francisco, October 2019)
- 2 workshop poster presentations at NeurIPS (Vancouver, December 2019)
- Kaggle Days Workshop (Tokyo, December 2019)
Rakuten Technology & Innovation Silver Award
(Tokyo, December 2019)
The Rakuten Technology & Innovation Awards are awarded to individuals or organizations which have brought innovation to society through advanced technologies.
Official Press Release:
Rakuten Announces Winners of Rakuten Technology & Innovation Awards 2019
Rakuten interview with MLT Founder Suzana Ilić:
Machine Learning Tokyo is democratizing AI
MLT is fortunate to have strong partners that support the organization and activities. MLT’s main partner for financial and organizational support is Mistletoe Japan Inc. – a Collective Impact Community with the mission to re-create a sustainable human-centered future using technology.
The community is made up of those who lead the forefront of the global startup movement including entrepreneurs, investors, researchers and visionaries, with the mutual goal to solve global social challenges that humanity will face in the near future. Mistletoe’s main activities range from startup support and investment, research & development, joint ventures to ecosystem development.
Furthermore, we are grateful for our many MLT event collaborations with companies like Google Japan, Microsoft Japan, Toyota Research Institute – Advanced Development, Mercari, Rakuten Institute of Technology, and many more. MLT also works together with Universities and Research Institutes in Japan, such as the University of Tokyo, the Earth-Life Science Institute (Tokyo Institute of Technology) and RIKEN CBS (Center for Brain Science) for Machine Learning lectures, workshops and bootcamps.
MLT is also an active part of the international tech ecosystem in Tokyo and frequently collaborates with diverse communities such as ON-1 Tokyo, Women Who Code Tokyo, Code Chrysalis, Le Wagon Tokyo, Tokyo Tech Startups, etc.
MLT Projects 2020/21
MLT has planned multiple open education, open source and open science projects, as well as academic and corporate collaborations for 2020/21. Donations, grants and sponsored resources will facilitate the sustainability of the nonprofit organization and support the following efforts in Artificial Intelligence:
- Production-level Machine Learning and Deep Learning – Application focused meetups on the deployment of Machine Learning and Deep Learning for different use cases and verticals, including projects in the field of AI for Social Good
- Edge AI – Research and deployment of Machine Learning and Deep Learning models on power-efficient edge devices
- Machine Learning Math & Optimization – Global remote sessions for engineers and researchers, strengthening the fundamental knowledge around mathematical principles and operations of neural networks
- AI Ethics – Interpretable and Safe AI – Fairness, Transparency and Explainability in Machine Learning.
- Diversity & Inclusion – Supporting women in Machine Learning and underrepresented groups in tech through events and corporate collaborations to provide visibility, community and resources.
The number of MLT’s community members continues to grow at a rapid pace, and aspires to have more than 8,000 members by the end of 2020. In addition to educational efforts with organizing workshops and talks, it strives to proceed with multiple AI projects, mainly focusing on social good. Growth is measured and evaluated via MLT members, Social Media Followers, academic and corporate partnerships and the impact of open source projects. With 15,000 followers on Social Media and a monthly outreach of 100,000 impressions only on Twitter, MLT has shown to be an extraordinarily active and engaged, fast-growing AI community that has become one of the most impactful community-driven initiatives in the world.
Supporting AI talent in Japan
Upon a government plan that has been laid out in March 2019 Nikkei reported that Japan will seek to increase the development of talent proficient in artificial intelligence to 250,000 people a year, up from just a few thousand today.
MLT will continue to support the AI ecosystem in Japan, providing a platform for individuals to deepen their technical skills and serve as an accelerator for the growth of an ecosystem, including companies, as well as academic and industry research labs. We will keep investing in open education, open source and open science efforts that will support and engage highly skilled talent in Japan and beyond.
Suzana Ilić is a Computational Linguist specialized in Natural Language Understanding for AI systems. She has received degrees in Applied Linguistics from the University of Innsbruck, with research on datasets, linguistic features and neural architectures for text-based affective computing. Research stays in Decision Intelligence and Deep Learning for NLP include the National Institute of Informatics Japan and the RIKEN Center for Brain Science. She previously held a contract position at Google Japan where she worked on NLU for the Google Assistant (Machine Intelligence Group).
As the Founder and Representative Director of MLT, Suzana has set the values and core principles of the organization, centered around open source, open education and open science. She has been leading open source machine learning and deep learning projects since 2017.
Yoovraj Shinde is a technologist and a software engineer.
He is the Co-Founder of MLT and one of the board members.Graduated as an Electronics and Telecommunication Engineer from Mumbai University (VJTI), he joined Rakuten as a software engineer and has been working in the e-commerce IT industry for a long time. He now works with Rakuten Institute of Technology as a technologist/prototype engineer for ML projects. He is also representative of Robot Club which is a group of DIY software and hardware engineers.
His responsibility for MLT as Co-Founder is to facilitate ways in which researchers and engineers can help other researchers and engineers to grow. He is in charge of Study Sessions to find ways to apply ML/AI technology (Low power Edge AI/ Robotics/ FastAI etc) for society.