Where to share amateur/unconventional code experiments...

Page 1 of 1 [ 2 posts ] 

biostructure
Veteran
Veteran

User avatar

Joined: 17 Dec 2006
Age: 38
Gender: Male
Posts: 1,455

04 Feb 2021, 3:49 am

I'd like to know if there are forums/sites, places on sites like GitHub where people like to share their personal hobbyist research code on "hard problems" like speech recognition/synthesis, face detection, augmented reality, etc. Not projects where people write new high level interfaces that call existing libraries/modules/Google APIs for these things, or re-implement some classic algorithm verbatim (those kind of codes are easy to find, and super boring to read), but where people just start fiddling with custom algorithms they pulled out of their you-know-where, often with more than a touch of arrogant grandiosity of "maybe I can do better" but ending in something really one of a kind and a great learning project, even if it falls (possibly far) short of a standard existing technology.

To give an analogy from another (though somewhat related) area, this guy made a clock out of only transistors and LEDs, and it looks really cool: http://techno-logic-art.com/tower.htm. Yes, there are tiny self-contained microchips that do the same thing, but using those you don't get to work with and see all the complexity of what is going on in there. I'm interested in doing the same thing in code--in a world where the standard response in the professional world to anything algorithmically hard is increasingly often to use some sort of "black box" like a neural network or SVM or complex pre-written math library, as a hobbyist I'd like to take a crack at making, for instance, a face detector built out of features I choose completely by hand as a mason builds a stone wall and use a lot of pencil-and-paper math to derive clever closed form approximations to how they move as the face rotates in space.

One barrier I find to doing this is my own mental health--I have a sort of manic-ish temperament which makes me very excited about such "overambitious" projects, yet also makes it very hard for me to actually sit down and write anything. I'm working on that, and that isn't something that reading someone else's GitHub or forum post can help me with. However, the other side of the equation is that I don't have a good place (except maybe here on WP) to announce if and when I DO get something like that working, and just as importantly to find other people's hobby projects along these lines to marvel at and be inspired by. I found that just finding out about that guy's transistor clock project really helped my sense of confidence that there are other people who actually appreciate this sort of thing--and that's not even an especially difficult circuit to design, even though building it I'm sure took a LOT of work.

How would I find these sort of "crazy" "stream of consciousness type" code projects out there? I've looked for things like "face detection from scratch", "Self-written speech synthesis", "manually built neural network", etc. and get nothing along the lines of what I'm looking for. One possibility I'm thinking is that maybe people who like to experiment with low-level, "math from scratch" type stuff tend to prefer stuff like inventing new fractals, simulating particle systems, etc. rather than things like object recognition because they look cooler and you can see the complexity in the end result--whereas if you have code that finds cats in videos, nobody using it can tell without looking at your code that you invented your own math vs. using some black box Google thing like everyone else. Maybe that's why it's so hard to find (in addition, of course, to the fact that it takes a "crazy genius" to attempt such a thing)?



biostructure
Veteran
Veteran

User avatar

Joined: 17 Dec 2006
Age: 38
Gender: Male
Posts: 1,455

11 May 2021, 2:09 pm

Anyone have any ideas about this?