/sci/ - Science & Math » Thread #13650876

47KiB, 628x628, 1626550893660.jpg

View Same Google iqdb SauceNAO

"Enhancing" an algorithm, how to do it?

Anonymous Fri 17 Sep 03:59:32 2021 No.13650876 View Reply Original Report

Quoted By: >>13652468 >>13653939 >>13653943 >>13655396 >>13655688

I'm a thirdoid, and our department tasked us with enhancing an algorithm for our thesis. I'm confused about what tractable topic is there for me to pick. Since the department here is equally clueless as well. I self-studied some maths from linear algebra, algebra, and real anal, so I guess I have the bare minimum for mathematical maturity, I can backread some stuff if I miss anything.

Anonymous

Anonymous Fri 17 Sep 2021 04:06:21 No.13650893 Report

Quoted By: >>13652441

Find an algorithm for a thoroughly niche mathematical problem.
Try to improve it by using data structures, chan's optimization technique, whatever else you can think of
If you can't improve it on every point, there is almost certainly a way to make a time-space trade off, then you can claim you improved it in terms of space or time
Just make sure the field is not actually well-researched and the state of the art algorithm is already optimal

Anonymous

Anonymous Fri 17 Sep 2021 15:49:38 No.13652441 Report

Quoted By:

>>13650893
Is the hidden subgroup problem niche enough of a problem?

Anonymous

Anonymous Fri 17 Sep 2021 15:57:36 No.13652468 Report

Quoted By:

>>13650876
Depends on your field but in software I often see people do stuff where they write a DSL library that allows you to easily run some obscure computations that nobody would ever want to do in slightly alternative way. They show comparative speedup between your library and the last time anyone cared to think about the thing in a paper from 2004 just by writing it with a modern compiler. Bonus points if you use a functional programming language where you can claim that the library automatically generates a proof of each computations correctness via the type checker or something like that.

Arno Niem

Arno Niem Fri 17 Sep 2021 22:34:18 No.13653918 Report

Quoted By:

Ever seen a program crash with the IT people having no clue why?

It's great fun. Especially when you ask them if they realized that there is a division in the formula that under certain conditions causes a DIVISION BY ZERO error.

Something basic like that can be quite an improvement - add error prevention.

Arno Niem

Arno Niem Fri 17 Sep 2021 22:45:24 No.13653939 Report

Quoted By:

>>13650876
Book suggestion;
https://www.springer.com/gp/book/9780387947464
it has the mathematically correct way to do stuff AND often a much simpler way to compute it (with precision etcetera indicated).

So that might ... hmmmh ... give you some inspiration.

Anonymous

View Same Google iqdb SauceNAO 1629298526287.jpg, 71KiB, 349x512

Anonymous Fri 17 Sep 2021 22:47:43 No.13653943 Report

Quoted By:

>>13650876
you have to give more information on your problem.
lots and lots of people do stupid shit because they don't know anything about the math they are using, even if highly optimized solutions have been well known for decades. it's very likely your problem has been encountered, if not in your field, then in the dress of something else.

so tell us more about what you are trying to solve

Anonymous

Anonymous Sat 18 Sep 2021 15:57:05 No.13655396 Report

Quoted By:

>>13650876
OK, an idea here, let me know what you think. You might get a patent out of it.

Topic: computational imaging for panoramic photography

Prior art: People use cell phone cameras to take a series of barely overlapping images and after 360 degrees the phone stitches these together into one image. The result isn't too shabby but could be better.

Improvement 1: Take instead a series of images, all overlapping at least 50 percent with the prior image and get at least double the images taken so that every pixel has two images to it. Then use computational super resolution to stitch together a panoramic image with better quality.

Improvement 2: use multiple main cameras to add extra details in areas of interest.

I have more ideas if this is interesting.

Anonymous

Anonymous Sat 18 Sep 2021 18:16:39 No.13655688 Report

Quoted By:

>>13650876
pick a problem from an engineering field that isn't know for being good at computers, but still tries doing computational work. I guarantee you'll get significant speed just because you know what the fuck you're doing

Anonymous

Anonymous Sat 18 Sep 2021 23:06:27 No.13656551 Report

Quoted By:

>8 posts, 7 posters
Seems OP abandoned the thread upon posting. We have been had.

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches