I'm a Computer Science student, specializing in Artifical Intelligence, who is currently nearing the end of his second last year of university. This means I'll have to do my Master's Thesis next year. However, we must already choose a topic this year.
My university offers quite a lot of topics, but none of them got me excited. None of them gave me the idea: "This seems fun, I really want to work a whole year on this", which is unfortunate. When you have to spend at least 700 estimated hours on something, it better gets you excited. It better be something you really want to work on.
There is a bright spot. We are allowed to propose a topic ourselves (which is not included in the list of topics). This may be in correspondence with a company. The idea to ask HN for a thesis proposal suddenly popped into my head, which leads to the following questions:
Do you have a thesis proposal for me? Or does your company have a thesis proposal for me?
If you can answer any of those questions with a "yes", feel free to reach out to me at hn_sebos [@] protonmail.com. Additional questions are always welcome.
One is the idea of separating multiple audio tracks. For example, if you have a MIDI file of two instruments, exported to WAV, it should be possible to train an AI to produce two WAV files that sound exactly like the individual instruments being played on their own. (Last time I heard a demo of this, it seemed like the AI hadn't built a good model for what the individual instruments should sound like, even though it should be possible to infinitely generate the training data).
The other is an AI for turning hand-written mathematical notation into a proper digital form (like LaTeX). Ideally you'd be able to give the AI a feed of a lecture, where notation was written on a chalkboard, and the output of the AI should be valid mathematical expressions. Bonus points if you can validate that the mathematical steps are correct (possibly by finding them in an existing corpus of mathematical steps).
DM (email) me if you want to learn more