According to Gartner’s 2018 CIO Survey, 50% of the 3,000 companies surveyed said they plan to apply AI to their operations, and 4% were already applying AI to their operations.

If there is a goal that humans want to achieve through machines, the machines must be trained to achieve that goal. In the past, when data was put into the machine, a program was created to produce output. With the development of AI, another method has emerged. When data and output are put into the machine, the machine creates a program that produces output. This process is called machine learning, and the program created by the machine is called a machine learning model.

Many people use the two concepts interchangeably, but machine learning is a subset of AI, and AI is also part of a workflow to achieve a goal. While there are AI-specific software like Caffe and Tensor Flow, there are also software like MATLAB that can be integrated into the overall workflow.

Although machine learning is now widely used, many engineers and scientists still believe that it is difficult for non-AI experts and inexperienced people to handle.
Misha Kim, MathWorks Customer Success Team Leader, said in her keynote speech at MATLAB Expo 2018 Korea on April 26, 2018, “You can apply AI to your work by using MATLAB even if you lack AI knowledge.” She also said, “Any engineer or scientist who has identified a problem and has a methodology to solve it can use AI,” and gave four cases where MATLAB helped apply AI to work.


Finding a suitable machine learning model with classification learner

The first is the case of a German PhD student in food science who created a machine learning model that can tell how crunchy a particular snack is. The student input data measuring the sound made when chewing the snack and the jaw force used to chew the snack into MATLAB and extracted features. And I classified the features using the Classification learner app included in the Machine Learning Toolbox at the bottom of the Apps tab at the top of MATLAB.
This app is designed to help non-experts who are new to machine learning to easily classify features, thereby reducing the effort involved in the classification task. It also allows you to instruct them to train all possible models with the classified features and view the results. Through this app, the student was able to find a machine learning model whose data output and actual results were almost identical. In this process, not a single line of code was written.


Achieving Deep Learning Effectiveness with MATLAB Production Server

The second is the case of Obayashi Corporation of Japan, which built a tunnel in Boston, USA. Obayashi Corporation used deep learning for effective drilling. Since Obayashi Corporation did not have the capacity to analyze the geology of the construction area, they decided to analyze 70 geological images of the construction area instead. Obayashi Corporation divided each of the 70 images into 15 parts, for a total of 1,050. Then, they labeled each image.

Obayashi tried to analyze the images with AlexNet, a deep learning-based image recognition software. However, AlexNet did not include geological knowledge, so it could not analyze the images well. So, they used transfer learning, which reuses existing similar models. If you automate MATLAB's transfer learning with MATLAB Production Server via the cloud, you can see the deep learning effect with just 5 lines of code. Obayashi was able to successfully complete the tunnel construction with optimized drilling.


Getting data with Simulink

The third is the assumption of predictive maintenance of wind turbines. In order to avoid failure of wind turbines, parts should be replaced according to the replacement period. However, if it has already failed, it is a waste in many ways to wait until a good wind turbine actually fails to obtain failure data. Using MathWorks' Simulink, you can obtain failure data much more easily and quickly. You don't have to write thousands of lines of code by hand. Simulink includes various project models. You can select and configure a wind turbine model with Simulink, input information such as the durability of parts, and then run a simulation. After the simulation is complete, you can obtain failure data to be used in the machine learning model.


Use it comprehensively

The fourth is the case of the auto parts company Denso. Toyota City, Japan, is a representative city that has implemented low-carbon policies. Denso wanted to find a way for low-carbon houses located in Toyota City to save as much as possible on electricity. The conditions given were as follows. The power source is solar panels and fuel cells, and electricity is stored in batteries. There are electric car chargers, but there is no information on when residents use electricity.
Denso used MATLAB and Simulink at every stage of the project. The project was completed in six months, and Denso said that it would have taken several years without MATLAB. It goes without saying that the AI utilization ability of Denso employees improved along with the project.


If you have the ability to identify problems and have a solution,

Misha Kim's keynote ended with a video. It was a video of a drummer who lost his right arm in an accident and received a robotic arm transplant, successfully completing a drum solo in a jazz concert with the help of AI. As such, AI is currently being used in many fields. Even non-AI experts and non-experts can apply AI to practical applications with the help of MATLAB if they have the ability to identify problems and come up with solutions.