The author also addresses the need for software engineering skills and suggests that those without such skills should start by implementing canonical data structures and algorithms. The author acknowledges that while there are many CS/coding skills that one might need to learn, these can often be picked up on the fly as they are less hierarchical than math. The roadmap is intended for those focusing on building models that are new in a theoretical sense, as opposed to deploying or scaling up existing models.
Key takeaways:
- The author provides a four-stage roadmap to go from high school math to cutting-edge machine learning (ML), which includes foundational math, classical machine learning, deep learning, and cutting-edge machine learning.
- Foundational math includes high school and university-level math that underpins machine learning, while classical machine learning involves coding basic regression and classification models.
- Deep learning involves multi-layer neural networks with many parameters, and cutting-edge machine learning includes transformers, LLMs, diffusion models, and other advanced techniques.
- The author emphasizes the importance of implementing key models from scratch for better understanding, and recommends resources like textbooks, online courses, and research papers for each stage of learning.