After spending a year working on Machine Learning projects, I finally decided to blog about my experience and perspectives that I have developed over the course of time. This is the first article from a series of upcoming articles discussing my work in the field of Biomedical Image Processing. I hope you all find it worth reading!

1. Learn Deep Learning, with a focus on computer vision and robotics: Given that you already have basic machine learning skills, I’d suggest following Andrej Karpathy’s computer vision deep learning course. It will teach you deep learning, with a focus on computer vision. If you prefer reading books, both http://deeplearningbook.org and Neural networks and deep learning are quite amazing.
2. Read research papers, watch technical talks: You can combine the goal of learning robotics, reinforcement learning while getting familiar with research papers. Deep learning changes fast, so for a lot of things you’ll have to rely on research papers. The Berkeley group is doing awesome stuff in Deep Learning + Robotics. Also, OpenAI is probably the best at deep learning + reinforcement learning.
3. Take up a project: There is a lot of dart art / black magic to deep learning. Nitty-gritties that you don’t pay attention to when watching a lecture, but which are crucial to get right for your model to learn anything. So ...

In an ideal world, a researcher with an idea could easily build on top of what has already been done (the 90%) and have 10% of work left to do in order to test his or her hypothesis. ... In practice, quite the opposite is happening. Researchers spend weeks re-doing data pre- and post-processing and re-implementing and debugging baseline models.

The difficulty of building upon other’s work is a major factor in determining what research is being done. The majority of researchers build on top of their own research, over and over again. ... I believe the main reason for this is that it’s easiest, from an experimental perspective, to build upon one’s own work. .... Baselines are already implemented in familiar code, evaluation is setup, related work is written up, and so on.

It also leads to less competition – nobody else has ...