This course is taught directly by the authors of Python Deep Learning PyTorch

Learn about the concept of artificial intelligence (AI) and deep learning, which is the basis of artificial intelligence in recent years.

| What is data science?

| What is artificial intelligence?

We will talk about the concept of artificial intelligence and talk about why deep learning has recently become popular. In addition, we will talk about where artificial intelligence is being used and how it is developing. We will talk about recent trends in artificial intelligence (GAN, reinforcement learning) and major issues.

| Multi Layer Perceptron (MLP)

Called the first artificial intelligence Perceptron and perceptron limitations, and MLP that overcame themLearn about. You can think of MLP as the basic structure of a neural network. MLP's learning algorithmI will let you know step by step. We talk about feed forward and back propagation and talk about advantages and disadvantages.

| Definition of deep learning

Learn intensively about what is the definition of deep learning and how it differs from a typical neural network. It deals with activation functions, dropout, and batch normalization that can mitigate the gradient vanishing/overfitting problem, which is a disadvantage of NN. Furthermore, we will discuss Auto-Encoder, which can not only classify, but also learn about new features.

| Convolutional Neural Network (CNN)

Looking at the history of deep learning, I think the most advanced model is probably the CNN model. It covers the CNN model, which began with image classification and has made tremendous advances. It talks about the characteristics of learning algorithms and their differences from typical NN. In addition, we will talk about various architectures (Resnet, Densenet), initialization, optimizer techniques, and transfer learning to improve CNN's performance.

| Recurrent Neural Network

Learn about the Recurrent Neural Network (RNN) deep learning model, which can reflect the characteristics of natural language. They learn the feeding process of the RNN model numerically, and the advanced Long Term Short Memory (LSTM) and Gated Recurrent Unit (GRU) models also explain the feeding process formally.

| Natural language processing tasks

There are many tasks in natural language processing. Among them, you will learn what kind of tasks are Tagging and Neural Machine Translation, which are selected as the most representative ones. In addition to specific examples for each task, a representative deep learning model structure for analysis methods is presented, and the weight feeding process of data is explained.

| ATTENTION

We present the limitations of the RNN model and introduce the Attention technique, which is a mechanism that has recently emerged in the field of natural language processing among methodologies to improve them. I will explain how each of Neural Machine Translation using Attention Mechanism and Tagging using Attention Mechanism can be applied. Also, Here is a brief introduction to Bert, who is currently focusing on research in the field of natural language processing.

| Generative Adversarial Network

Most of the artificial intelligence we talk about these days uses deep learning models. Typical machine learning or deep learning models ended with classification and regression.

However, the advent of GAN brought about such a big paradigm that it is not an exaggeration to say that the development of artificial intelligence was one step ahead. At that time (4-5 years ago), generating data beyond classifying and predicting was unthinkable even then (4-5 years ago). With the advent of GAN, along with reinforcement learning (the basic principle of AlphaGo), it has become a field that cannot be removed from artificial intelligence.

The picture above is a fake image created by GAN, which has the best performance as of a year ago, that doesn't actually exist in this world. Currently, a more advanced model has come out.

| CycleGAN (GAN that became the basic model for Style Transfer)

[Beyond generating data] We have begun to develop into various fields using GAN generation principles. Among them, the representative model is CycleGAN, which is a Style Transfer model. They change a picture like a photograph or a picture like a picture, change day and night, and change the seasons.

CycleGAN is a GAN model that interchanges the two domains of an image in this way. This CycleGAN has become the basic base line model for GAN utilizing Style Transfer.

| CAN (GAN model that creates works of art)

Generating data doesn't create anything new because it's created within the learning data after all. So it's far from art. This is because if you create it within learning data, it's just an “imitation” after all.

The CAN model creates art by slightly changing GAN's learning principles. It is said that they surveyed humans and obtained scores similar to actual works of art.

| Various fields where GAN is applications/developed

Other than that, GAN has developed in a wide variety of fields.

We will briefly introduce various GANs, such as Radial GAN for generating structured data for machine learning rather than images, DeGaN, which is a model of how to generate diverse, high-quality images in situations where there is little learning data, MGAN that combines models of multiple GANs, and SRGAN, which changes low image quality to high definition.

| What should I do to study DS/AI?

For those studying DS/AI for the first timeI'll give you one piece of advice.

- R vs Python

- A major for AI?

- Is it possible to be a liberal arts student?

- Is graduate school an essential option?

- AI-related jobs and required competencies