Artificial Intelligence with Python

This course is designed for learners who want to understand the fundamental principles of deep learning from a mathematical perspective. Before using complex deep learning frameworks or libraries, the goal is to systematically understand the principles by implementing the operating principles of neural networks directly through matrix operations and NumPy .

First, the structure of the neural network and the learning mechanism are explained by dividing them into the forward propagation and backpropagation stages. We will focus on matrix multiplication to explain how the input is transformed in each layer, how the difference between the output and the correct answer is calculated according to the loss function, and how this difference is backpropagated through the network to modify the weights. Through this, you can see with your own eyes how learning actually takes place.

The loss function is a key factor that determines the learning direction of a deep learning model. In this lecture, we will implement the formulas directly from MSE (Mean Squared Error) , BCE (Binary Cross Entropy) , and CCE (Categorical Cross Entropy) , and explain through specific examples which problem types each is suitable for. Through this, you can gain deep insight into why you should choose this loss function, rather than simply using code.

In the activation function section, we implemented major functions such as Sigmoid , ReLU , Tanh , and Softmax one by one, and designed it so that you can intuitively understand the characteristics, pros and cons of the functions by drawing graphs. For example, you will learn by visually analyzing why ReLU is more effective in deep neural networks and how Softmax creates probability distributions.

Based on these components, we will start with a simple 1-layer or 2-layer neural network , implement a multilayer perceptron (MLP), and practice solving a classification problem using real data. You can also directly adjust hyperparameters such as learning rate, epoch, and batch processing to experience changes in learning curves and model performance.

This course is not just a simple theory course. It is designed to help you learn 'why' and 'how' neural networks learn by explaining how each line of code is connected to the formula and implementing all the formulas by converting them into matrix operations based on NumPy .

Ultimately, through this course, students will understand the inner workings of deep learning frameworks (PyTorch, TensorFlow, etc.) before using them, and if necessary, will be able to implement neural networks themselves. The experience of seeing deep learning, which seemed complicated, simplified into a single mathematical system and working right before your eyes will provide learners with great confidence and insight.

This course is the perfect choice for those who want a balanced learning experience between theory and practice, intuition and mathematics.