Is Machine Learning a Chloroblastic Topic?
When I first heard the term Machine Learning (ML), I was mystified. Could it be that ML is optimization? This confusion was echoed by many, as nobody could offer a satisfactory explanation. This is because ML is inherently intertwined with optimization, yet it is more art than science. It is about making decisions on what to optimize, how to optimize it, and what the results mean.
The Immaturity of Machine Learning
Unlike mathematics and physics, which have rich classical literatures still useful today, the field of ML is not yet mature. There are no traditional blueprints to follow as in structural design. However, there are tons of empirical observations and experiments to be conducted. While ML does have some theoretical foundations in areas such as numerical optimization, probability, statistics, differential calculus, and linear algebra, these are merely prerequisites for the real stuff in ML.
Getting Started with Machine Learning
The best way to begin your journey in ML is to develop a personal project. Start small with something like face detection, object detection, or speech recognition. This will give you a solid foundation as you implement things from scratch, such as the Backpropagation algorithm, to fully understand them. Later on, you can work on open-source ML projects to further enrich your skills.
Consuming Quality Information
Stay informed about the latest developments in machine learning by reading journals from major research labs like Google DeepMind, FAIR, Baidu, Microsoft, and OpenAI. Some journals can be quite technical, but as you get used to ML jargon, it will become easier. Watching videos can also be very helpful; freely available tutorials on YouTube are a great starting point. Focus on the basics first, then move on to talks from Google, Microsoft, or Facebook on AI. These videos can also help you understand the broader applications of ML, such as self-driving cars and pedestrian detection.
Engaging with the Community
Stay up to date by reading and answering questions on platforms like Quora. Engaging with the community can provide you with valuable feedback, helping you gain a deeper understanding of ML. You can also form a self-study group with friends to monitor each other's progress. This collaborative approach can be very beneficial.
A Field in Flux
Machine Learning is a rapidly evolving field. A lot of the literature from the 1990s is out of date because during that period, Deep Learning (DL) was not the state-of-the-art. At that time, ML was heavily based on linear regression, support vector machines (SVM), and some neural networks (NN). The internet was not as big as today, so big data did not exist back then. The processing power available at that time, particularly GPU technology, was still in its infancy.
With such rapid changes, there is no clear roadmap in ML. It is a young field, and only with time will it mature to the point where we can write textbooks or ebooks with a longer shelf life. In 10 years’ time, ML will likely be quite different from what it is today. This is why online sources may often present conflicting information; it is a field that requires real personal effort to understand and contribute to.
ML is still a kid, in the sense that kids learn and grow. So don’t get frustrated that you find different information online. By putting in the effort and digging into your own insights, you can start to unlock the many possibilities in ML. Remember, as you have your own ideas, such as experimenting with A, D, and K instead of A, B, and C, you might achieve something entirely unique. The key is to keep experimenting and implementing your ideas quickly.
In conclusion, ML is a challenging and rewarding field that requires both theoretical knowledge and practical application. By embracing it with a curious and learning mindset, you can navigate the chloroblastic territory of ML successfully.