About machine learning programming

Recently, I've been studying AI and machine learning using Python, and it's quite interesting to start doing it. Especially, I think it's really convenient that you can easily use TensorFlow and Keras with just a browser using Google Colaboratory.

At first, I'm a little confused about where to start this kind of study, but there are many books on AI and machine learning, and when I read them at a bookstore and somehow keep in mind the keywords, I somehow understand the image. After that, if you make an online reservation for a book that seems good from the library and read it, your understanding will gradually deepen.

It was a book called "Deep Learning Mechanism to Learn While Working with TensorFlow and Keras" that was very interesting to me and deepened my understanding of machine learning. As a result, I was able to grasp the concept of logistic regression, softmax function, sigmoid function, ReLU, convolutional filter, CNN, etc. in addition to matrix calculation. Since I had studied a neural network around 1990, I was always wondering what was different between that and deep learning, but when I look at CNN, I can understand that you can raise the accuracy rate by combining quite various techniques.

And, recently I bought a book called "How to make AI/machine learning / deep learning apps with Python", downloaded the code from Github to Google Drive, and then touched the code from a laptop computer at home or at a coffee shop. I'm studying a little bit about how the concrete implementation looks like this. In particular, I just read the concrete coding by Keras last week, and I was surprised that a machine learning program could be easily realized just by declaring the settings, just like a Lego block (or Scratch). Certainly, I think that a machine learning program can be created without mathematical understanding, but even so, it seems to be good to proceed with Keras after understanding the basic idea that realizes this.