Nanalyze

Top-10 Artificial Intelligence Startups in Japan

The Land of the Rising Sun is a peculiar mix of tradition and modernity. Nowhere else in the world can one see centuries-old shrines sitting comfortably next to high tech skyscrapers in such harmony. Nowhere in the world does the airport ground crew stop what they’re doing so they can wave goodbye to departing planes until they’re out of sight. Nowhere in the world will you find customer service that drips with genuine sweetness with no tips expected. Nowhere in the world will you find a people as endearing as the Japanese.

Japan has been in isolation from the rest of the world until its Meiji period (1868-1912), and its late opening has resulted in a unique combination of Eastern and Western values. While youngsters party beneath the neon shades of the largest metropolis in the world, the country still retains a conservative approach to life. If you’re a foreigner, expect to be viewed with curiosity, and if someone turns a cold shoulder when you ask for directions, it’s probably just because they’re a bit scared of you. And then you’ll come across a cultural phenomenon like Babymetal, and realize there’s no way you will ever come close to understanding this fascinating country.

On the tech front, Japan is considered the leader in the field of robotics, but perhaps a lesser known fact is that they were the second nation after Canada to adopt a national AI strategy, something they released in March of 2017. The strategy focuses primarily on AI applications for productivity, health, and mobility. We scoured Crunchbase to find the most funded AI startups in the Japanese ecosystem as of today.

Name Application City Funding
(USD millions)
Preferred Networks Internet of Things Tokyo       130.0
ABEJA Big data Tokyo         45.4
Ascent Robotics Autonomous vehicles Tokyo         17.9
Cinnamon Optical character recognition (OCR) Tokyo         17.0
LeapMind Edge devices Tokyo         13.5
Cogent Labs OCR, natural language understanding Tokyo         12.9
Moneytree Fintech Tokyo         10.5
MUJIN Inc. Robotics Tokyo           7.0
Alpaca Fintech Tokyo           6.7
MJI Smart assistant Tokyo           5.0

Click for company websiteFounded in 2014, Tokyo startup Preferred Networks has raised $130 million from the likes of Toyota, Hitachi, and Fanuc to apply machine learning to Internet of Things (IoT) use cases. Their latest funding round put a $2 billion price tag on the four-year-old startup making it Japan’s only unicorn according to the experts over at CB Insights. The startup’s only public products are an automated manga coloring tool and some opensource developer materials for neural network frameworks. The company’s core activities center around autonomous vehicles, machine learning for robots and machine tools, and medical diagnostics.

Preferred Networks is relatively tight-lipped about what they’re up to, and that’s no surprise given that they’ve become an outsourced AI research lab for some major-league partners like Toyota, Fanuc, and the National Cancer Center of Japan. Preferred Networks also launched one of the most powerful private sector supercomputers in Japan last year in partnership with NTT Communications Corp. As Bloomberg rightfully puts it, “what sets Preferred Networks apart from the hundreds of other AI startups is its ties to Japan’s manufacturing might.”

Click for company websiteFounded in 2012, Tokyo startup Abeja has raised $45.4 million from a list of investors that include Google and Nvidia to develop big data analytics based on IoT sensor data. The company has built an analytics platform that can be integrated into virtually any business with enough data to be crunched for producing valuable insights. Abeja is also offering specific products optimized for use in retail, manufacturing, and infrastructure on the back of their core platform.

The structure and data flow of Abeja's AI platform

The structure and data flow of Abeja’s AI platform – Credit: Abeja

The retail solution analyzes in-store customer behavior to optimize business and inventory management. The manufacturing module improves production efficiency by automating product inspection and predicting machinery failures. The infrastructure offering prevents breakdowns by detecting possible malfunctions and scheduling maintenance, something we talked about recently. Abeja is now planning to expand to foreign markets like China and Thailand with localized versions of its analytics service.

Click for company websiteFounded in 2016, Tokyo startup Ascent Robotics has raised $17.9 million to develop software for autonomous cars and industrial robots. Ascent’s approach is built on their AI learning architecture called Atlas. The company says it is the “skeleton” on which AI training simulations work and is equipped with models generating real-life environments, situations, and feedback in a virtual setting, basically teaching AI algorithms how to behave. With the mix of real and simulated data, the AI’s learning efficiency is reported to be 50 times higher compared to when only real-life data is being used. Ascent wants to create a fully autonomous (level 4) vehicle software by late 2020, after which the company is reportedly planning an IPO. The startup has Ken Kutaragi, former Sony Computer Entertainment CEO and creator of the PlayStation on its board, and is planning a new investment round of $30-50 million in the coming months.

Founded in 2012, Tokyo startup Cinnamon has taken in a total of $17 million with their latest round of $6 million closing just days ago. Covered earlier in our article on “7 Startups Using AI for Robotic Process Automation,” Cinnamon offers a smart scanner algorithm with Optical Character Recognition (OCR) that can read and understand text and handwriting. The Flax Scanner can process virtually any type of document in a matter of seconds with an accuracy of 99.2%, saving up administrative and support staff for more value-added activities. The company has offices in Japan and Vietnam, and is going to use its latest funding round to expand into the U.S.

Click for company websiteFounded in 2012, Tokyo startup LeapMind has raised $13.5 million in funding from the likes of Intel to develop embedded deep learning solutions on edge devices. Edge computing means that computations are performed on distributed devices like smartphones or sensors, and not in a centralized cloud environment. This removes the need for cloud connectivity, provides optimized performance, and allows for quicker communication with local communications hubs called nodes.

The steps of creating and deploying a deep learning model by LeapMind

The steps of creating and deploying a deep learning model – Credit: LeapMind

LeapMind’s Delta suite of products provides the hardware, the data labeling and training framework, and the deep learning model builder to create and deploy deep learning models on edge devices without background knowledge in programming.

The company claims their model builder reduces the three-month process of model design, training, compression, and conversion to just one day. Use cases include automated inspections on food production lines, identifying construction anomalies using drone or fixed-point camera images, and analyzing market trends based on online image analytics.

Click for company websiteFounded in 2015, Tokyo startup Cogent Labs has raised $12.9 million to develop AI solutions for natural language understanding, character recognition, and time series forecasting. The company’s Tegaki Optical Character Recognition (OCR) app targets the most common job being replaced by AI – the file clerk – and has a handwriting recognition rate of 99.22%, similar to Cinnamon above. It is only available for Japanese language, and Cogent has signed up mammoth conglomerate Softbank as one of its clients.

Features of Cogent's natural language understanding algorithms

Features of Cogent’s natural language understanding algorithms – Credit: Cogent Labs

The Kaidoku natural language understanding engine uncovers insights from text-based data like news, social media streams, and documents. It provides overview, filter, and search functions, as well as visualizations across time. The company’s time-series forecasting algorithms analyze large amounts of historical data to come up with long- and short-term forecasts in situations where traditional statistical models usually fail. This is useful for applications such as securities trading, and the startup’s client list includes one of the country’s biggest investment banks, Nomura.

Click for company websiteFounded in 2012, Tokyo startup Moneytree has raised $10.5 million to develop a financial data aggregation platform for individual consumers and corporates. Users can register bank accounts, credit cards, and other accounts like securities holding accounts or loyalty cards, and Moneytree pulls their transactions and assets into its platform automatically. Data is saved for the lifetime of the membership, providing valuable historical information as well as current analytics.

How Moneytree's MT Link for financial service providers works

How Moneytree’s MT Link for service providers works – Credit: Moneytree

The company also offers an automated corporate expense registration and reimbursement service working along the same principles, and an integration service for financial companies called MT Link. The platform currently covers 2,600 service providers across the country.

Click for company websiteFounded in 2011, Tokyo startup Mujin has raised $7 million in funding to develop AI-based motion control software for industrial robots. While robots are traditionally programmed to perform certain tasks or movements, Mujin’s controllers allow robots to “think” through their movements without any pre-programming, adjusting to reality as they go along, much like most humans do.

The company doesn’t build robots but provides the controllers which are compatible with most of the robots being manufactured today. Mujin’s solutions are mainly used in logistics, warehousing, and factory automation for picking, packing, and sorting tasks. The startup has developed the world’s first warehouse without any human staff in partnership with JD.com, the Chinese e-commerce powerhouse. Seems like it’s only a matter of time before some of the biggest warehousing robot manufacturers will look to implement Mujin’s advanced motion control technology into their machines.

Click for company websiteThis next startup we’ve come across before twice; first in our article on 6 Startups Using AI for Algorithmic Trading Strategies and second in our article on Free Algo Trading for Tech Savvy Traders. Founded in 2016, Tokyo Startup Alpaca has raised $6.7 million to develop predictive models for financial markets and offer the world’s first free algo trading platform. The startup’s AlpacaForecast AI Prediction Matrix uses AI to calculate short-term price forecasts for certain currency pairs, equities, and fixed income instruments. The matrix comes in the form of a dashboard and plugs into everyone’s favorite financial data provider, Bloomberg. The broader AlpacaForecast framework is built with ultra-high speed data storage technology, designed by Alpaca from scratch, and optimized for financial time series. The company is working with regional and local banks like Jibun Bank and MUFG Bank, the largest bank in Japan.

Founded in 2015, Tokyo startup MJI (which stands for More Joyful Innovation) has raised $5 million to develop a virtual assistant called Tapia. The tool looks like a kawaii version of Google’s virtual assistant or Amazon’s Alexa and offers all the necessary functionalities you’d expect. It has voice and facial recognition capabilities, and is able to make calls, organize your schedule, read news, and play music. It also acts as a safety monitoring device you can use to check on your loved ones requiring assistance.

Credit: MJI

The startup has released a development toolkit for companies wanting to expand Tapia’s original functionality as well. MJI has also developed a home medical assistant called “anco,” based on Tapia’s hardware, and in collaboration with the NTT group. “anco” can gather and share information from vital signs reading equipment like thermometers and blood pressure monitors, diagnose patients based on real conversation, and provide remote support from nurses. Sounds like maybe they should partner with another firm we looked at called Qolty which is using big data to improve medical studies.

Conclusion

Many of these startups are developing generic AI technologies applicable to many use cases across many industries. This ties in well with the national AI strategy that envisions a whole ecosystem of AI algorithms speaking to each other by 2030 – something you’d expect from a collective society.

If we happened to “miss” your Japanese AI startup because you were too busy playing pachinko to update Crunchbase, we know you won’t say anything because you are far too polite to complain. Just drop us a note instead and the next time we’re in Tokyo – the town where all the cool startups seem to hang out – we’ll take you to Omoide Yokocho and you can tell us all about what you’re getting up to.

Worried that AI might steal your job? You should be. Why not get a Masters in Machine Learning from London's Imperial College, one of the top-10 universities in the world? And you can study online!

Intelligence that’s not artificial

Let our MBAs stimulate you with insights once a week.