After having finished my master's thesis, I finally found the time to setup this blog and start working on the posts that came to my mind during the years. This post is an attempt to outline the direction of this blog and its intended audience.
My entry into the field of Machine Learning was my work on the Battlesnake team at my university, where I implemented several Reinforcement Learning algorithms. Thus, many topics that I want to cover have their application in this area. I'm not going to write introductions to Deep Q-Networks or Policy Gradient methods, as there are plenty good ones on the web already. Instead I want to capture ideas that changed my perception of a concept and deepened my understanding of a topic. Additionally, I want to collect short reviews of books that I read and whose contents I found interesting.
For scientific education, a bottom-up approach is the best way to learn the right mindset and tools to solve the problems of a field. There are many great books that follow this style and complement the lectures at the universities or enable us to learn about a topic by ourselves. However, I think that blog posts are often more useful when written in a top-down manner. When looking for a post about a specific topic, I search for it and begin to skim the posts that I find.
Let's say I want to find a post about some novel application of Neural Networks. If a result of my search starts by deriving the chain-rule of differentiation, I either have to find the passage in the text, where it gets interesting for me or discard the text right away. Top-down posts, on the other hand, pick their readers up where they are, when they have a topic in mind that they want to learn about. The readers can read the post as far as they want, but the essence is right at the beginning. This is also beneficial for the short time intervals that our modern work places allow to investigate a new concept or topic.
If you want to get in touch with me, you can find my contact infos in the about section.