A question for data scientists about Deep Learning Thesis I am writing because I would like to ask more experienced people in Deep Learning about the (possible) topic of my master's thesis. I am not sure if it is not too complex and even possible to finish in 1 year for a student who has no professional background in deep learning and just finish a few related courses on Coursera and I need to ask your opinion. To put it in a nutshell: The topic title is: De novo peptide sequencing from tandem MS data. The result of my work should look like this: https://github.com/nh2tran/DeepNovo (but I need to figure out a different idea because that research group have a different kind of data). I would not have any 'technical' support because the tutor has only background in biochemistry. I start having doubts if it is a doable thesis. I can still change it. Thank you for the answers!
 A: Your question may well be closed as "opinion-based", but I will give it a shot.
Deep Learning is a statistical learning tool. If you have some background in statistics (and programming, and the scientific background), then getting to grips with it in a year is doable. (Realistically, you would need to take less than a year, since you will likely need to dig through the relevant scientific literature as well, and write up the thesis.)
Many people are "thrown in at the deep end", especially in industry, and told to learn DL in a few weeks to put together a DL-based tool. Most of these would be very glad indeed to have a full year!
My misgivings are less about the time frame, and more about people using DL without a strong statistical background. I do not think anyone really understands a straightforward Ordinary Least Squares linear regression without at least some grounding in statistics, probability theory, probability distributions and linear algebra. DL is far more complex than OLS (although one can consider it as just a very complex nonlinear regression). There is a very real danger of getting overconfident and misusing the tool. From what I have seen online, most tutorials spend far more time and effort on the computational/programming aspects, and very little on the statistical intuition which is IMO at least as important. Note that this applies both to your situation and to the one of that hypothetical business DL learner. And that there are many, many users in exactly your situation, of (probably) not having a strong statistical background but learning to use very powerful tools anyway. Just as when putting a power drill into the hands of someone who has never used a hammer or a screwdriver, the results may be gory.
Then again, here at CrossValidated, we have a larger population of statisticians who do have that background. So you can certainly discount my feelings as the grumblings of an old curmudgeon who is afraid of modern technology. (I don't think that is the case, but YMMV.)
Bottom line: I do think you would be able to put together a thesis within a year, although it might not pass muster with an experienced statistician or data scientist.
