How to come up with substantial research projects?
I've been learning about deep learning for several months, first by reading through Understanding Deep Learning then by implementing papers touching on CNNs, residual connections, transformers, gpt models, vision transformers, etc.
My original plan was to learn by reproducing papers and running small experiments, but I've found this approach somewhat unfulfilling. Most implementations end up being only a couple hundred lines of model and training code, and I often struggle to figure out what to do beyond simply reproducing the paper and running some basic experiments.
For those of you who have spent a lot of time working on deep learning projects, how do you come up with interesting, long-term project ideas? How do you go from understanding the fundamentals and reproducing existing work to building something more substantial that keeps you engaged and challenges you to learn new things? I have a few ideas and papers that I'm interested in exploring, but I don't want to fall into the same habit.