Hacker News new | ask | show | jobs
by llm_trw 842 days ago
I swear the only thing that the Bolte Bridge attracts more of than seagulls is photographers.

Also the desktop I have now for ML work is powerful enough to do a full 3d simulation of a lens using the full Maxwell equations from first principles. I remember doing the back of the envelope calculations that I'd need the Bluegene/L back in my undergrad days to do it for real, well: https://bnnbreaking.com/arts/video-gaming/nvidia-geforce-rtx...

What a time to be alive.

3 comments

> I swear the only thing that the Bolte Bridge attracts more of than seagulls is photographers.

Thanks for encouraging me to search about it, and quickly realize Bolte Bridge is a natural target due to relative level of architectural sophistication as well as being easily accessible.

https://en.wikipedia.org/wiki/Bolte_Bridge#/media/File%3ABol...

Why would you use ML to develop well understood formulas from first principle, if you can also have a properly developed program for the same purpose?
I think the ML that the desktop is for, and the optical calculations, are 2 unrelated projects.
Yes, OP was doing linear ray tracing to see how the lens would perform on the video. I remember doing the same in a computational physics class way back in the day and talking to the lecturer about what you'd need to simulate real optics without approximations in the equations.

The answer was the worlds fastest super computer at the time. I was a bit shocked that he claimed that was the state of the art for lens manufacture in industry too - no idea how true that was. But figured that if you needed that much computation it made sense.

Well I have 2x that under my desk now and I use it to do local development before I push to cloud machines with 100x the power where the actual work happens.

Today most commercial lens design software still uses geometric ray tracing. (Even if some can do EM simulations too.)

Not entirely surprising. You want to be able to run stuff a bunch of times to optimize it.

https://www.synopsys.com/content/dam/synopsys/optical-soluti...

Which would actually make a lot of sense, now that I think about it...
You could develop novel lens shapes, for one.
May you please expand? Truly interested