Hu Yoshida

Deep Learning with Hitachi AI Technology/H

Blog Post created by Hu Yoshida on Mar 14, 2017

AI stands for Artificial Intelligence, computer systems that can achieve intelligent activities like those of human brains such as learning, reasoning, and judgment. The learning method employed by early AI systems was to have computers learn the rules and logically find solutions based on the rules. As such, the systems were limited in that they only found solutions within the scope of what they had learned.

hitachi AI.jpg

More recent types of AI being used today are designed based on hypotheses of data that people would enter, implemented as programs, and analyzed and developed for a specific application. One disadvantage of being based on hypotheses envisioned by people is that they usually do not produce results that surpass human ideas and are not general purpose.


One well known AI system is IBM’s Watson, which is described by Wikipedia as a question-answering computer system capable of answering questions posed in natural language. Watson was named after the first CEO of IBM, Thomas J. Watson. In 2011, Watson competed on the quiz program ”Jeopardy” and defeated two former Jeopardy winners. The IBM team provided Watson with millions of documents, including dictionaries, encyclopedias, and other reference material that it could use to build its knowledge. All the content was stored in RAM and Watson could process 500 gigabytes, the equivalent of a million books, per second. Watson's main innovation was its ability to quickly execute hundreds of proven language analysis algorithms simultaneously to find the correct answer.


Hitachi has developed similar technology that can analyze large volumes of text data that are subject to debate and present reasons and grounds for either affirmative or negative opinions on these issues in English. By using multiple viewpoints, it is able to present reasons toward a single perspective. One use case was in medical diagnosis. Recently this was expanded to the Japanese language.


Hitachi uses natural language processing in the Hitachi Visualization Predictive Crime Analytics (PCA), to ingest streams of sensor and Internet data from a wide variety of sources like weather, social media, proximity to schools, subway stations, gunshot sensors, 911 calls, etc and crunches all this information in order to find patterns that humans would miss. Police investigators build crime prediction models based on their experience with certain variables like slang words that may come up on Twitter and assign a weight to each variable. Hitachi’s PCA doesn’t require humans to figure out what variables to use and how to weight them. You just feed it those data sets, and PCA uses machine learning that decides over a couple of weeks if there is a correlation that could predict a crime.


A new learning method is called "deep learning," which incorporates the mechanisms of neural circuits in the brain. Like neural circuits, the computers themselves learn the characteristics of the data entered into them for learning, they can make judgments even about patterns that they have not yet learned. Using this method, it is possible to automatically produce explanations for images and motion pictures, conduct highly-accurate automatic translation, and make forecasts in a variety of fields including financial markets, weather, and professional sports.


Hitachi AI Technology/H (here after “H”) is the name for Hitachi’s Artificial Intelligence engine that uses deep learning and is one of the key technologies in the Hitachi Lumada IoT platform. H was announced by Hitachi in 2015 to respond to a wide range of applications. H can learn from voluminous amounts of data and make judgments on its own, eliminating the need for people to set up hypotheses in advance and finding solutions that humans had not conceived.


In the attached video a robot outfitted with H is placed on a swing made of toy blocks. The purpose of the experiment is to maximize the swing amplitude without providing prior knowledge on how to do so.  The robot can bend and extend its knees but has no knowledge of how to swing. At first the movements are random and the swing barely moves. The robot starts to move the swing in less than 1 minute and in 5 minutes has come up with a swinging motion that exceeds human conception. This is truly an example of deep learning where human input is not required.



H is designed for general purpose AI. H connects to existing systems to learn from different kinds of data and grow according to the situation. Just as in the swing example, H searches for factors and conditions relevant to business objectives from among the various type of data that is already there then searches for a method to optimize these objectives


Hitachi has been involved in R&D and Proof of Concept (PoC) activities targeting "Hitachi AI Technology/H" for about 10 years, and has established an extensive track record of business reforms in a variety of fields, including finance, transportation, distribution, logistics, plants, manufacturing, and healthcare. It has been applied in 57 projects in 14 areas.


Hitachi AI H use cases.png


The key to H is its general purpose nature. According to Hitachi Corporate Chief Scientist, Kazuo Yano, “H does not require customization nor tuning for the warehouse or the store. Under changing business situations, the system learns from the situation, and the changes.”


H will be a powerful component of our core IoT Lumada platform.