[Background] Promoting on-site process innovation using AI

At Panasonic Connected Solutions Company, the Gemba Process Division, which was launched in January 2019, is one of the core organizations for promoting Gemba process innovation that solves customer issues. In the manufacturing, logistics, and retail industries, goods, information, and people are intricately intertwined, and sophisticated solutions are required to solve the problems faced by the workplace. In particular, the AI/IoT Solution Department of the headquarters, which introduced Driverless AI this time, said, "For the miscellaneous areas that are difficult for customers to enter, we will directly enter the site to sort out problems and lead to problem solving. We provide AI solutions,” explains Masataka Ejima, chief engineer of the department.

[Issues] Desire for AutoML that is easy to visualize as well as accurate

In machine learning in AI, the process is largely divided into the areas of data set preparation and model creation. It is possible to improve the accuracy of “Like other companies, we are currently spending a lot of time on the front part. We wanted to secure enough time to assess the situation,” explains Takashi Atsushi Tajima, chief engineer of the department. What attracted attention as a solution was a machine learning automation tool called AutoML (Automated Machine Learning), which can automatically generate models.

Mr. Tajima explains that what is required of AutoML is not only the accuracy of the algorithm used, but also the ability to visualize the factors that greatly affect the results. “For example, when making time-series forecasts, we have to make sure that the model does not use future information that cannot be used in actual operation, such as the weather on that day. What is needed is something that can be visualized in an easy-to-understand manner and which factors are effective.”

Especially in the case of time-series forecasting, the occurrence of Leakage may produce better results than the original performance, and there are cases where the expected effect cannot be obtained even if it is applied to the field, so it is difficult to determine which variables are used. becomes important. Of course, it is necessary to explain the results derived by AutoML to the customer, and transparency such as what algorithms are used and what variables are used is one of the important factors.

[Selection] Tools that enable advanced visualization and the presence of a partner that serves as a bridge are significant

In fact, the company has been using AutoML tools for some time, but was looking for a solution that would reduce the burden as much as possible, would be more transparent, and would be able to do the same or more than the existing AutoML.It is said that he encountered Driverless AI provided by Nika Networks. "We've tried several products, some don't match the accuracy we need, some algorithms...

You can download the continuation of the case here.

Customer story 1st page sample