Beomjun Kim's Blog

I tried to understand the CNN in terms of math, but it was not easy to organize its concept explicitly, especially because of the concept 'convolution'. However, thankfully I could seize its intuitive understanding with 3B1B's video . Attention was pretty not easy to analyze mathematically, so I had to skip the math-hard comprehension. But still Hyukpenheim's video  is surprisingly useful. So I only organized the notes about RNN and LSTM. Download PDF

 In this session, we inquired into backpropagation. The Korean Youtuber '혁펜하임'(@hyukppen)'s videos were very useful, and comprehensive. So although things about SGD, Batch GD, and Mnibatch GD are written referring to other sources, the part for the principle of Backpropagation is actually just a lecture note for the video. Download PDF

 Last Saturday, the first session of the BITAmin was held. I was happy as I could meet many students who have similar interests to me, and also they are passionate. However, the club focuses on more practical things, and also the backgrounds of members are various, so the contents of the sessions are intuitive. However, I'm also interested in the mathematical principles of ML and DL, so I'm determined to organize the math used in the sessions. So I've posted this article. The last session handled about introduction to deep learning, loss function, activation function, and gradient descent. Download PDF

Yesterday, I attended the OT of the Data Science academic club 'BITAmin'. (Instagram: @bitamin_official) The club won a favor of me. As the club has a clear aim, I could feel we are on the same wavelength. There, I met many students who had the same, or similar purpose as me. It was almost the first time that I met those people. Anyway, I'd like to review the interview for participation in the club. Just before having the interview, we needed to have a coding test. But it is not about algorithms. It tested the participants whether they could handle basic Python, pandas, and numpy. Fortunately, the operation team uploaded the past exams, so I could prepare for the test well. However, the interview was still a big problem. It was a 3:3 style, but the other interviewees have had project experiences. So I told my enthusiasm and passion. And I appealed that although I have no experience, at least I definitely understand the things that I studied. In fact, when I was asked a tech...

My university set all of the students majoring in economics to take this course. This course deals with not only basic stats knowledge like conditional probability, binomial/discrete/continuous distribution, and Poisson / exponential distribution but also hypothesis tests, like fundamental concepts of the test, t-test, paired t-test, variance estimation using Chi-square distribution, and F-distribution. Unfortunately, the opportunity to conduct real statistical analysis as it was a very large lecture. But the contents were pretty dense. So it was very helpful when I studied linear regression and ANOVA myself.

Macroeconomics! Though microeconomists are more than macroeconomists, to be honest, most people usually think of macroeconomics when they are asked to imagine 'economy'. I'm not an exception. So I studied this course with a pretty big interest. AD-LRAS-SRAS graph, NCO-NX, Phillips curve, and long-run Phillips curve... all of them are fascinating topics. However, as I'm a Korean, NCO was the most interesting. South Korea was sort of the most famous country for rapid economic growth. During the speedy growth, I(investment) was not enough as I  = S for a closed economy. Of course, for rapid growth, the saving was insufficient. So using I = S - NCO, at that time the trade deficit was a big problem. But for this trade deficit, NCO < 0, so I > S. It must have been an important factor in the growth economy. Usually, the trade deficit is regarded as a considerable social problem. But sometimes it's essential...