Google's Machine Learning Event in March Google Developers ML Summit Tokyo vol 2 I’ve been participating in this event, so I would like to share my learning.

What is Machine Learning?

Machine learning can analyze a large amount of data for a given problem, and lead to an answer “without having humans teach you how to solve it.
In the past, a human engineer had to teach his computer how to solve an issue in order for him or her to answer it.In contrast, machine learning can analyze large amounts of data and calculate it more accurately than humans think.
It is already used in a variety of applications, such as image recognition to determine faces and natural translation by computer.

In this seminar, an example of a problem that determines whether the person is an adult or a child from data about his body was explained.

in the case of traditional programming

An algorithm that considers an adult to be more than cm in height or kg in weight is transmitted and determined by the machine.
The downside of this method is that if you have a child who has great growth, it will be misjudged.

In machine learning,

Data about adults, data about children are inputted and analyzed in large quantities from there to automatically deduce the law that forms a boundary between kids and adults.
The accuracy of the judgment exceeds 99%.

Artificial Intelligence, Machine Learning and Deep Learning

Machine learning is often spoken in the context of artificial intelligence, and I think that it is a reason why machine can calculate automatically even if an algorithm is not taught to human beings.
However, it’s important to note that machine learning can only answer “a question with the right answers” (see a picture to determine if you are a child or an adult) and not provide answers for questions without the right answers like how we should live.
Deep learning is a method of machine learning that replicates the model in which human brain learns something on computer.
The theory itself has been around for a long time, but thanks to the recent improvement in computer performance it became possible to implement at a practical level and attracted attention again.

Google's Machine Learning Solutions

Google has provided a number of solutions that make it easy to use machine learning, and there was also an introduction at the seminar.
The higher the degree of freedom and difficulty from above, the lower you go down, the lower the degree of freedom and the less difficult it becomes.

tensor flow

• This is the de facto standard solution for machine learning right now.
• The creation of a learning mechanism called the Learning Model is more difficult in data science than engineering, but it seemed easy to use another person’s learning model.
• There’s also a js version, which is easy to use on the web, so you can easily determine emotions from images in your webcam or see what’s in an uploaded image.

Google Cloud AI

• Tenor flow is more of a product set for specific applications such as speech recognition and image recognition, with plenty of freedom.
• You can take advantage of the massive amount of data that Google has accumulated so far to use tuned services in the cloud.
• There were a lot of APIs that would be interesting to play with, such as video recognition, image recognition (recognition of faces and text included in images), translation, natural language analysis (reading emotions from sentences, etc. )

ML Kit for Firebase

• It’s the best option if you want to quickly add machine learning-powered features like barcode, facial recognition and text recognition into your mobile app.
• If you want to customize that is not provided in the API, it will be implemented in conjunction with the learning model of tensor flow lite.

That's the report, but it was a run.
It was a very intellectually curious day, and it made me want to make something interesting with machine learning!