Friday, February 23, 2024
HomeTrendingNon-digital human resources can also use machine learning | What is required...

Non-digital human resources can also use machine learning | What is required for future no-code tools!

In recent years (machine learning), major IT companies have been actively acquiring no-code platforms, such as Google’s acquisition of AppSheet. In addition, Amazon has released a no-code application development service “Amazon Honeycode”. Japanese vendor companies are also riding the trend of no-code/low-code development, and tools that can improve operational efficiency, build websites, and develop applications are being released one after another.

machine learning

While there are many new tools and cloud environments, DATAFLUCT Co., Ltd. aims to ” democratize data utilization ” , realizing no-code from data connection to machine learning model construction, implementation, and operation that can also utilize non-digital human resources. We develop and provide end-to-end machine learning platform ” Comler “.

What are the characteristics of Comler?

This time, we interviewed Mr. Hayato Kumemura, Representative Director of DATAFLUCT Co., Ltd. and Mr. Kazuki Harada, CTO, about Comler’s features and future requirements for no-code tools.

Table of Contents

  • Work on new business and product development
    • Self-introduction
    • Key points when starting a new business
  • “Comler” for end-to-end support
    • Enjoy the world of machine learning
  • Emphasis on data quality and renewal
  • Future prospects
  • Lastly

Work on new business and product development

self-introduction

ーーPlease introduce yourself.

Mr. Kumemura: I started my career in the new business department of a company, and have launched more than 15 new businesses at multiple companies such as Macromill, Recruit, and Nihon Keizai Shimbun. While creating new businesses, I wanted to create a company that fused digital and new businesses, so I started DATAFLUCT about three years ago.

At DATAFLUCT, we are engaged in consulting and launching new businesses related to data utilization business on a daily basis.

Mr. Harada: I joined a major system integrator and have been in charge of cutting-edge fields such as IoT, AI, and big data. Utilizing the skills he acquired there, he has participated in many hackathons and won the highest award.

Since I liked cloud technology from that time, I changed to Microsoft and implemented architecture in various areas such as infrastructure and applications as a cloud architect. While gaining experience at Microsoft, I thought it would be a waste not to create a product even though I have development skills, so I joined DATAFLUCT as a business consignment in January 2020, and joined DATAFLUCT as CTO in November of the same year. After being in charge of creating platform products, I am currently developing products that are right in the middle of our company’s goal of “democratizing data utilization.”

Key points when starting a new business

ーーPlease tell us the secret of new business and AI business.

Mr. Kumemura: The concepts I cherish are ” future orientation ” and ” long-term orientation .”

Future-oriented means thinking about what it would be like 100 years from now, and backcasting what needs to be done now and what problems need to be solved to realize it.

Long-term orientation is the idea of ​​creating businesses from goals, such as the SDGs, which set 17 goals to be achieved by 2030 .

Combining these two is the heart of business development. When looking at the time axis of 10 years or 100 years from now, our stance is that the business will spread to the world in about three years. Therefore, we use our imagination to see what will be popular and what will attract attention in that era, and work backwards from the future to develop our business.

When launching a new business, DATAFLUCT’s style is not to create what customers want now, but to dive into the times while making hypotheses about what the world will need in the future. At DATAFLUCT, we are developing businesses in fields that will become commonplace in 3-5 years and that competitors cannot develop.

ーーDATAFLUCT takes a stance that is not self-sufficient, but is that stance important?

Mr. Kumemura: Yes. We believe that it is not the person who develops the algorithm that is amazing, but the person who transforms the algorithm so that it can be used in business . For example, it is not the person who invented the phone that is amazing, but Steve Jobs who designed a product like the iPhone and made it portable.

We believe that it is important for business to focus on issues and design instead of developing technology.

If Microsoft and Google have developed technology, I think it would be better to use those technologies and think of products that will make us and Microsoft and Google profitable.

“Comler” for end-to-end support

DATAFLUCT develops and provides a machine learning platform “Comler” that allows non-data and IT specialists to connect data, build, implement and operate machine learning models with no code.

ーーPlease tell us about the characteristics of “Comler”.

Mr. Harada: One of Comler’s features is that it incorporates elements such as no-code, templates, and automation into each function so that machine learning can be used by people other than data and IT specialists .

In the field of AutoML 1 , where DataRobot pioneered the market, management by big vendors will increase in the future. With the release of various no-code/low-code services, we thought that providing an environment that could be used more easily throughout the company would be important for the operation after introduction.

It’s simple to use. Select ‘Quick Start’ and then either import a file from local or register as a dataset. In the future, we would like to add options such as Google BigQuery and Amazon S3.

Then select the target column and choose which cloud to use. After that, simply select your analysis method, set up your model, and run.

You can also see the evaluation of the developed model, so you can judge whether it is the best model by comparing the numerical values ​​​​of each score.

After developing a model, it needs to be deployed in order to be included in an app, but it can also be tested after implementation. Currently, this part is made for engineers, so if you have any questions, you can send a “request” and DATAFLUCT will reply with the results.

Enjoy the world of machine learning

ーーPlease tell us how you came to develop Comler.

Mr. Kumemura: The background behind creating Comler’s predecessor, “DATAFLUCT cloud terminal.” According to the “Current Status and Issues of Digital Data Utilization” in the “2020 White Paper on Information and Communications” released by the Ministry of Internal Affairs and Communications, only 11.9% of domestic companies use artificial intelligence (AI) such as machine learning and deep learning. We have not been able to realize our predictions.

Considering that there is a bias, I think that it is actually lower than 11.9%, so we are aiming to improve that figure to 20% and 30% in the future.

There are many reasons why deep learning and machine learning projects fail, such as  <strong>cannot prepare data</strong> '' ,cannot access data ” ,  <strong>cannot understand data '', andcannot find data scientists ” . Elements like these keep us from achieving our intended purpose and bog us down.

Mr. Harada: In addition to not being able to use convenient tools made overseas, Japan’s IT literacy is low compared to other countries, which is a persistent problem. As for overseas tools and new tools, there are few people who have prior knowledge, so it is necessary to adjust the required level and develop tools that can be mastered gradually.

I believe that by growing both users and services, people will be able to enjoy the world of machine learning.

ーーWhat trends do you see in cloud models?

Mr. Harada: Since Google Cloud Platform (GCP) is based on deep learning, the accuracy will increase if the amount of data is large, such as 1000 lines or more. If the amount of data is 1000 rows or less, the accuracy is higher in Azure etc.

Azure is versatile and I think there are many cases where it can be used, but there are still many things we don’t understand, so we would like to verify it in the future.

Emphasis on data quality and renewal

ーーIn addition to differences in accuracy between models, I think that “data quality” is also important. What do you think?

Mr. Harada: When I listened to clients who used the service “cloud terminal” before the renewal, there were many cases where they were frustrated because they could not prepare the data . If we can prepare well-organized data, we can create and deploy models with AutoML, but we learned that many companies cannot prepare well-organized data. 

Therefore, in “Comler”, we have updated the functions of the previous stage of model building and strengthened the “data connection” and “data standardization” parts. When organizing data, we look at outliers and the difference between the minimum and maximum values ​​in the data, but there is a general rule of thumb. We have also adopted automation technology so that data can be processed in a few steps.

In addition to the function that can simulate what kind of results can be obtained when actually using the model, the model accuracy maintenance and monitoring function, and the data pipeline / MLOps pipeline are introduced so that highly accurate forecasts can always be made.

We also added an automation function for process reproduction by . In addition, we have added a simulation experiment notebook function to promote communication between internal business users, data scientists, and decision makers who actually use Comler, promoting collaboration within the organization and making the project proceed more smoothly. you will be able to

▼ Is the Algorithm Era Over? If you want to know more about data quality, which will be important in the future, click here

Future prospects

ーーPlease tell us about your future prospects.

Mr. Kumemura: The true purpose of spreading no-code tools is to bring about change from within the company. It is said to be a so-called DX-like transformation, but it is more like an OS transformation. OS is the way of thinking, the logic of decision making, how to proceed with work, the sense of value against risk, tolerance for failure, speed, etc.

In the process of thinking about how to change them all at once, I thought that the learning cost would be the lowest if I could create my own algorithms and be able to implement them. So, by delivering tools that change the OS in these organizations, we hope to help develop people, change the OS of the organization, and change the business.

Mr. Harada: Every time I look at the chaos map of image analysis or video analysis, I hope to be able to select good things more quickly. For example, if you use AWS for image analysis because you usually use AWS, I think Microsoft may actually have higher accuracy.

There are things that you can’t know unless you try them one by one, so I want to make it easier for users to find the best tools.

Lastly

DATAFLUCT is working on the development and provision of Comler, advocating the democratization of data utilization. Comler incorporates elements such as no-code, templates, and automation in each function so that even non-digital personnel can utilize it. By accurately grasping market forecasts and client usage, and improving from cloud terminal to Comler, you are making steady progress toward realizing your vision.

There is concern about the shortage of data scientists in Japan, but with the introduction and utilization of no-code tools within organizations, even non-digital human resources will be able to smoothly proceed with projects that utilize data.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Recent Posts

Most Popular

Recent Comments