Courage from an offline classroom

I got the courage to start teaching in Class 101 in an offline classroom. In the LG Electronics DX School course, I met students with different majors and experiences while teaching 400-hour deep learning classes across multiple riders. Overall student ratings Average score of 4.5 out of 5It was, and the respondent's 92.3 percentgave it a score of 4 or 5. In particular, it was a great help when students from literature and entertainment who were unfamiliar with mathematics and development talked about understanding the difficult flow of deep learning.

Of course, praise wasn't the only thing that made this course. There were also honest opinions that “the faster you progress, the easier it is to lose the big picture in the middle” and “even if you think you know when listening to explanations, it becomes overwhelming again when the formula and model come out all at once.” I immediately incorporated that feedback into the next lesson. Before showing the structure of the model, I first talked about why such a structure was needed, and before calculating the formula, I wanted to first show what problems the formula solves. This course is my answer to the possibilities and regrets I've seen in an offline classroom.

In the age of artificial intelligence, what we really need is not a “correct answer,” but a “good question.”

In the future, we will do more work with artificial intelligence. Artificial intelligence examines numerous data in a short time, suggests various solutions, and sometimes shows combinations that humans have never thought of. However, as artificial intelligence becomes more powerful, the role of humans does not disappear. Rather, it becomes sharper.

Deciding what to look at as a problem to solve, dividing a big problem into smaller problems that can be handled by artificial intelligence, comparing the basis and limitations of the answers proposed by the model, and deciding what to choose from various possibilities and being responsible for the results. All of this is still up to people.

The best starting point for me to understand this role The history of deep learning I think it's inside.

Why “history”

The history of deep learning is not a series of model names and performance records. It discovered unsolvable problems, examined why existing methods failed, and found new solutions by changing the way they looked at problems A person's storyThis is it.

When the problem was data that couldn't be divided into a straight line, researchers piled up multiple judgments.
As the layers deepened, it became difficult to understand which connections to fix and how much, and I found a way to convey the wrong answers backwards.
When the problem of forgetting the first part of a long sentence appeared, I made myself choose what to remember and what to forget.
As it became difficult to capture everything in one memory, I came up with a way to revisit the original text when needed.

One solution always opens up new possibilities, and at the same time leaves problems to be solved next. This trend contains attitudes that can be learned even when we solve problems with artificial intelligence today. The attitude of dividing the problem into smaller pieces, the attitude of asking again about conditions taken for granted, and the attitude of turning failure into information for the next attemptThis is it. This course not only conveys deep learning knowledge, but also talks about the problem-solving process of researchers who changed the times.

The reason for adding “culture” and “line drawing” to the title

Dare to say the title of this lecture 'Liberal education' I added the word Liberal arts is not a shallow reduction in expertise. What we must have together to understand a strange world and make our own judgments Common languageThis is it. I believe knowing what deep learning can and cannot do, what to entrust to artificial intelligence, and what judgments humans should abide by is a necessary culture and common sense for all of us living with artificial intelligence.

“Draw a line” is a common picture in this course for understanding complex deep learning models. The model divides what is similar from what is different from the data, and learns the criteria for distinguishing between answers that are appropriate and those that are not. The shape and method of drawing lines have changed from era to era, but what kind of data is divided and what to look at as a good answer has always been an important question. There are always people at the beginning and end of setting those standards.

What I want to leave to all the students

You don't have to be someone who memorized a lot of model names when you are stubborn. Instead, I hope to be able to ask this when I meet a new artificial intelligence.

“What problems did this model change to solve? Which of the answers should I choose, and who is responsible for the results?”

I hope this course will be a small compass to help people collaborate with artificial intelligence through good questions and responsible choices instead of vaguely fearing or unconditionally believing in artificial intelligence.

Essential Liberal Arts in the AI Era: Understanding without knowledge of mathematics or coding <Liberal Arts Deep Learning>

Essential Liberal Arts in the AI Era: Understanding without knowledge of mathematics or coding <Liberal Arts Deep Learning>