This is a short guide on how to approach a problem with no known solution from a PhD angle. I wanted to keep it here as reference. It was posted on reddit by user /TillWinter on /r/MachineLearning.
Define the state of knowledge, tools, understanding you already have. Like a table with grades, a simple list what every you feel comfortable. Make it as clear and simple.
Define your goals; what is it, you want to have understood. PhD means to walk the border of the unknown, so what are the questions you have that needs to be answered
Put this 2 pieces of paper in front of you, with a third one in between. Your job now is to find the “shortest path for the accomplishment of your task”. Meaning using the a high abstraction, what is needed to solve your question.
In most cases you cant find a simple path just on the first try. If so you are, as we say, “drilling a thin board”. I mostly start with the “solution side”, so what is the step to be taken before you reach your goal (also try to define it abstract). And than the next and so on. Try to put a “node” to each side (start point and goal) till you let them meet. This is now your shortest path.
Now look at your steps, you will see in your minds eye a list of dependency, missing knowledge, something you need to work out. That are your milestones to learn, experiment and analyse.
Now try to estimate the time frames you need to accomplish each sub-task, add them up and write them under each step. Now multiply that number with a factor between 3-20. This is your real esteemed now. Best way is to use hours as unit.
Now high efficient work per day is max about 2-4 hours. And for most people 2-3 days a week. Rest is, Posters, emails, calls, talking, drinking coffee and so on. Never try to overcompensate by forcing yourself to work more than 20 hours highly effective per week. You will burn out. Learning is a highly effective task.(If you actually want to learn the subject you are working on)
Now calculate all the time you might have and compare it with the workload you expect. If your task/project is to big, make it smaller. Till your median workload you can accomplish is in the limits of your Project estimate.
Now show it to a colleague, ask him about his estimate of your task without revealing yours. If it fits, all is good.
Every step contains sub-sub-task for you, try to plan them out. What is to study, what is it that others have done. Do I need tools I don’t know yet, do I have to build them.
Write your own personal monthly schedule. Set a side at least 20% learning and reading time.
Check your speed of work and try to stay in the margin
Now in the process you will find new questions, new thinks you need to know. This will “thicken” your path. There are 2 kinds, the “must have” and the “nice to have”, say good bye to the nice to have or do it in your free time as hobby. Sleep and reevaluate your “must haves” 90% of the time its a “nice to have”.
You are finished with your smallest path before the time is up? Now you can feed it.(if it is a 3 year fellowship, you are in the beginning/mid of your 2nd year now) Try other smaller different ways between your steps. Write paper about it. Feed it till you have about 9 month left.
Clean up! tie up your lose ends. Make a nice poster and write a last paper. Teach your the generation following you what you have learned, even just for fun. Its a great way to train speaking in front of people and its fun with undergrads.