(Sorry for the following lengthy text and abstract/non-concrete phrases because I can’t explain my complex ideas, which are much more lengthy, here)
I have great ideas to speed-up many transformer models.
(I am not being arrogant when I use the word arrogant, It is reality my ideas are great)
Let me talk about me and you will understand why I am doing this when I talk about the interaction of machine learning and data structures.
I am a undergraduate student of mechanical engineer but I have attended a course of data structures in the computer science faculty in my university.
(I have decided to study a little bit of computer science to become a better professional and not to became an entrepreneur and against all expectations I am sure I am able to create a disruptive startup. I was not planning to become a entrepeneur at the beginning of the my undergraduation…)
I studied some unusual and new data structures that are able to speed-up transformers by roughly 100 times !!! without using expensive GPUs.
(I have studied data structures because I was planning to became a better researcher in mechanical engineering and not a researcher in computer science…)
The 100 times speedup sounds unlikely to happen and I will explain it better later when requested but this speedup is indeed feasible.
My ideas will make many transformers economically viable at really large scale but I need help to modify some models and upload other models at huggingface.co because I am not a computer scientist and I can’t do everything alone.
I am sure transformers are not implemented at large scale like the web because researches/entrepreneurs are not using the cleverest tools.
There are many improvements in terms of precision over the vanilla transformer but they are insanely expensive and to my knowledge almost them use tools of machine learning and not from other areas of computer science.
As for me the cleverest tools to make transformers economically viable (in terms of speed and memory usage) are not inside the machine learning field but in the data structures field.
(as I said in the beginning, some data structures are able to do a lot !!! for transformers and I am sure this is an unexplored connection between two knowledge areas, despite many many articles published about transformers).
The main problem is:
I live in hardship at Brazil and I can’t afford even to pay $9 for a pro plan (at least at this moment, maybe in future).
There are other problems: are I don’t have a diplom, I haven’t published any article in my life, I don’t have a github account so many readers will think I am a scammer/con man. But I assure you: I am not asking for money or investor at the beginning of my startup. I am asking for some intelligent machine learning scientists and other computer science cientists to help me.
Where/ to whom should I ask for help to create a startup?
Thanks in advance