What I've observed is that to really get to the deep learning, the deep knowledge, requires moving into the 'elite', and learning from them.
Think of it as applied Pareto: find the top 20 who do 80% of the work, and learn from them so you can move into them.
One of the most absolutely valuable pieces of 'research' I have done as part of my Master's is learning about systems that were designed and built before 1980. Then, learning about how many of those ideas were recycled and or forgotten in the trash heap behind Unix's house.
Think of it as applied Pareto: find the top 20 who do 80% of the work, and learn from them so you can move into them.
One of the most absolutely valuable pieces of 'research' I have done as part of my Master's is learning about systems that were designed and built before 1980. Then, learning about how many of those ideas were recycled and or forgotten in the trash heap behind Unix's house.