2019: shaders, dreamwidth, and more

[anhinga_anhinga]

I hope for us all to have a creative and safe New Year.

Computer art news: I started to play with OpenGL shaders and with Shadertoy.com: https://dmm.dreamwidth.org/20076.html

Comments

[misha_b]

Your post about mental inertia is right on target. I do not know why but it seems particularly strong in ML (not just neural networks).

Happy New Year!

[anhinga_anhinga]

> mental inertia

да, интересно, можно ли по этому поводу что-то сделать (в смысле организации какого-то сопротивления этому феномену), или надо просто принять к сведению, что вот так всё оно и есть (and just persevere) :-(

(probably all machine learning has this property which neural nets possess: relatively simple math + lack of "hidden parts of nature"; this makes effects of mental inertia very visible and obvious, while in other areas there are mathematical or experimental complications, so the same effect might exist, but be less obviously visible.)

С Новым Годом!

[misha_b]

Yes, that seems plausible. The math is not too hard + the experiments can be easily repeated in a computer.

I am not sure what to do about that. I am concerned about the level of conformity and peer pressure we observe in ML (and, likely, other areas as well). I think the incentives are aligned to push researchers to produce large quantity of average results. This tendency has really accelerated in recent years.

The funding agencies similarly tend to be mostly interested in large collaborative projects which are not likely to shake the existing status quo (despite claiming to support transformative research).

My personal feeling is that giving more money to individual researchers on a less competitive basis rather than concentrating on large projects would be a step in the right direction.

[anhinga_anhinga]

Yes, these trends and observations apply across the board (they span all fields of research).