Research: My two cents

Short-term Machine Learning Research Projects

Get familiar with the prior work done in the domain especially in context of project scope. Do the literature survery. Be it research papers, blogs, lectures, etc. Read it. Try to replicate the results, if possible. A good paper on reproduciblity
Use version control. Things will break and undo button will be handy.
Formalize your approach and technical thinking by writing extensively. This will bring clarity about your objective, hypothesis and the ways to actually get feedback on what really works and what doesn’t work. Your supervisor can help you better if they can trace your work and the rational behind technical decisions. So writing it all down really helps.
Irrespective of your findings and results of your experiments, discuss it with your team and guide to know what you should do differently in next iteration. Imposter Syndrome may kick in early if you are stuck on a problem and not discussing it with your supervisors or peers.
Reuse the standard structures prevelant in research projects. Investing in systematic approach early might slow you down a bit but will save weeks and months of time because of easy detection of improvement scope in your project. [Read this chapter]
Avoid overcomplication. Start with a simple solution which works and do rapid iterations to improve it further instead of a really complex solution with slow iterations. Think in terms of minimal number of components required to solve the problem to desired level of performance.
Get a thorough understanding of your evaluation metric. Why this metric? How does it correlate with what our objective?. It will help with the optimization (hyper-parameter tuning, etc)

Short-term Machine Learning Research Projects

Long-term Research

Deep Reinforcement Learning

Blogs worth slow reading

ArXiv Paper