I went as a Monbusho scholar in the period 2004 - 2009 and this is not the case anymore. Nowadays you spend a lot of time filling forms and getting access to basic services like cellphones or Internet. They do give you around ¥160K these days...
My startup has a fast nearest neighbor engine that allows to have far matches: http://simmachines.com/Products/r01.html
We hope to be the "berkeley DB" of the big data era.
Feedback would be greatly appreciated!
This is a very important data structure from a historical point of view, but unfortunately it doesn't work well when you are dealing with complex spaces. Try LSH as mentioned before or something that is 10X faster: http://simmachines.com
You can solve complex things with k-nearest neighbor as long as you use an appropriate distance function. This is the beauty, a distance function abstracts the complexity. I tackled a tricky biology problem by applying a cascade of similarity filters. Check the presentation I gave at the Clojure/conj 2011:
http://prezi.com/zaaoq6pjrl2z/clojure-conj-final/
My startup provides a very fast similarity engine (in a DB of 100 million objects I can find similar objects in under 20 millisec. with one CPU) in case you worry about scalability. URL: http://simmachines.com
The SISAP metric spaces library comes with an implementation: http://sisap.org/Metric_Space_Library.html
You can also compare BK-trees with other methods and you may encounter that their performance is not so great, specially for high dimensional spaces.
VP trees are well known but not necessarily the best approach for large datasets or high dimensional objects. I am creating a startup that offers the fastest similarity search engine built so far (10X faster than LSH from MIT). See more here: http://simmachines.com
There was a talk around where they were using two replacements at most(or two hash tables), and two hash functions. At insert time you add the object to the first hash table, if there is an object in the bucket, you then try with the second hash table. If you follow this approach a very large percentage of the buckets is going to be full. The remaining objects set (those that cannot be inserted in the first two hash tables because the buckets are full) is so small that you can keep them in a list or in a "standard" hash table. At some point I had a secondary storage implementation of this approach and it was good indeed.
If your dimensionality is low then I totally agree, there is a library that can help you to achieve this: http://code.google.com/p/uzaygezen/
If you are dealing with objects with intrinsic dimension >=20 then you should use other techniques.
When I was building OBsearch.net I used this book extensively. I grant you that some explanations are not so detailed. On the other hand, the explanations of the most relevant techniques are quite good. Check for instance the section that explains LSH or the explanation about Navarro's SAT.
If you count the number of people killed by wars in Europe or North America in the 20th century, Latin America is the place to be. Latin America has numerous social problems but the number of conflicts (and their magnitude) between countries has been traditionally very small.