Xerox and JBig2.
On the scale of things too horrible to contemplate, “document-altering scanner” is right up there with “flesh-eating bacteria.”
Nice analysis on word2vec and how it is a more complicated version of matrix factorization problem. here
Inverted-index can use trie or hash lookup. I would think trie would be easier for looking at similar terms (just hash the new term again), but it takes up O(26) for each pointer on each level, so O(26^n)
Picking up similarities in forest decay, etc, in the midst of noise. landtrendr.pdf
Another simplified walking controller , by Adrien Treuille in his PhD. http://grail.cs.washington.edu/projects/graph-optimal-control/
A really cool walking simulator / animation tool that allows you to continuously move between different parameters in motion capture data. What's interesting is how they implement it! (eigenwalkers ) eigenwalkers.pdftrojechapter07b.pdf and wdp2002_troje.pdf
Dynamic_programming is very helpful! Basically, DP computes most all of the possibilities and picks the best one?? (very inefficient-sounding) I thought there were overlapping subproblems though…
Use the details of the problem space to your advantage. For example, if you want to get from point A to point B <far away> on a map, you only need to use the interstates for most of it (you don't have to compute for all of the roads in the middle), and getting the average distance of the houses in a city is fine for most of it.
Or just don't have them…YouTube Link! https://www.youtube.com/watch?v=cMx_W5sWC78
One company that does it is called Rhythm Engineering, and they implemented a system in Hillsboro OR along Cornell Road (report shows ~20 seconds reduction in travel time along the whole length, done by traffic engineering firm Kittleson and Associates). $48,000 per intersection.
It only makes sense for 10's of millions of files. still not sure why this limit
Normal behavior is it should be fast.
Runtimes and space usage of all algorithms and all operations.