A few months ago, I decided to self-learn machine learning and AI as a personal challenge.

I am Mac, African and student in civil engineering.

With a strong passion for mathematics, I turned my attention to computer science in January 2025. I first mastered computing logic before learning C programming.

By July, I felt the need to move past small-scale projects and specialize. Although systems programming caught my eye, my studies don't leave me enough time. However, with all the buzz around AI on social media and in the news lately, I’ve started exploring the field.

While researching the subject, I realized that I already possessed the basic prerequisites, such as linear algebra and some probability and statistics. That’s when I decided to dive in.

I’ve been reading books and watching YouTube videos to learn. I’m learning the most from MIT’s videos—I'm really grateful for their OpenCourseWare program and fully support their initiative. I started with machine learning, and after grasping the theory, I decided to move on to some hands-on practice.

I realized that to make things easier, I needed to use a high-level language like Python. After doing some research, I decided for educational purposes to code my own functions from scratch without using frameworks like PyTorch. It was tough at first—especially with backpropagation—but I managed to succeed. I've completed several projects, including: a model to predict Coronavirus probability (training error: 0.1, test error: 0.1), a house price prediction tool, and a project predicting wind turbine power output (training error: 0.002, test error: 0.003). I also worked on other projects where accuracy was limited by a lack of data.

As I dived into deep learning, I started to realize that the term 'Artificial Intelligence' is a bit of an exaggeration, since it's essentially a matter of mathematics and ingenuity. I am now studying neural networks and CNNs (Convolutional Neural Networks), and I’ve already completed a few projects in this area.

Following the course's path, I moved on to reinforcement learning, starting with Markov Decision Processes. I also built a small game project, even though it was quite basic. Afterwards, I dived into Recurrent Neural Networks and built a project focused on word prediction based on that architecture.I also have to deal with hardware limitations, since my computer lacks the processing power needed.

I know there is still so much to discover—like decision trees, random forests, unsupervised learning, generative models,... But at this stage, I am fascinated by everything I’ve learned. I used to think it was something extraordinary, but I now realize it’s 'just' math and ingenuity. Now, whenever I hear the term 'Artificial Intelligence,' I find it quite exaggerated (and I suspect experts in the field feel the same way).

If you have any updates on new algorithms or AI trends, please share them in the comments. I'm always eager to learn more!